[python] Use UUIDs for LeRobot dataset association keys

docs/docs/pypaimon/lerobot.md

@@ -46,22 +46,18 @@ with row tracking, no primary key, and `bucket=-1`. The indices below identify
 LeRobot records; they are independent of Paimon's internal row IDs. These
 relationships are used by the reader and do not declare database foreign keys.
 
-The diagram and field tables below describe **list imports**, including a
-one-element list such as `source=["/data/source_a"]`. This form adds
-`dataset_index` to frames and Episodes and also supports later appends.
-[Single-source shorthand](#single-source-shorthand) has a smaller schema.
+The same three-table schema is used for one source and for a list of sources.
 
-[![LeRobot metadata files map to the Dataset table, Episode Parquet maps to the Episode table, and frame Parquet plus media map to the Frame table. Dataset and Episode indices relate the three tables.](/img/lerobot-data-model.svg)](/img/lerobot-data-model.svg)
+[![LeRobot metadata files map to the Dataset table, Episode Parquet maps to the Episode table, and frame Parquet plus media map to the Frame table. Dataset IDs and Episode indices relate the three tables.](/img/lerobot-data-model.svg)](/img/lerobot-data-model.svg)
 
 ### Dataset fields
 
 `robot_data_datasets` has one row for each source, even when multiple sources
-have the same label. Source ordinals follow the input order; appends allocate
-new ordinals after the existing sources.
+have the same label. UUIDv7 values retain the order of serial imports and appends.
 
 | Field | Paimon type | Meaning |
 | --- | --- | --- |
-| `index` | `BIGINT` | Source ordinal, starting at 0. Referenced by frame and Episode `dataset_index`. |
+| `id` | `STRING` | UUIDv7 assigned to this source; referenced by frame and Episode `dataset_id`. |
 | `name` | `STRING`, nullable | Optional label supplied through `dataset_name`; omitted names are null. |
 | `info` | `STRING` | JSON containing the source's original `meta/info.json`, including `features`, `fps`, robot type, and counts. FPS is not a separate column. |
 | `tasks` | `ARRAY<ROW<task_index BIGINT, task STRING>>` | Source-local task IDs and their text from `meta/tasks.parquet`. |
@@ -81,7 +77,7 @@ forced to `BIGINT`.
 
 | Field | Representation | Meaning |
 | --- | --- | --- |
-| `dataset_index` | `BIGINT` | Source ordinal referencing `_datasets.index`. |
+| `dataset_id` | `STRING` | Source UUID referencing `_datasets.id`. |
 | `episode_index` | Source integer type | Episode number within that source, starting at 0. |
 | `dataset_from_index` | Source integer type | Inclusive start of the Episode's source-local frame `index` range. |
 | `dataset_to_index` | Source integer type | Exclusive end of that frame range. |
@@ -90,7 +86,7 @@ forced to `BIGINT`.
 | `meta`, `data`, `videos`, `stats` | `VARIANT` columns | Grouped source fields describing metadata locations, data shards, per-camera video locations/timestamps, and per-Episode feature statistics. Empty groups are `{}`. |
 | Other source fields | Source type, or a `VARIANT` for a `/` prefix group | Custom Episode metadata, such as quality labels or camera calibration. |
 
-For list imports, a field such as `data/chunk_index` becomes the
+A field such as `data/chunk_index` becomes the
 `"chunk_index"` key inside `_episodes.data`. A field such as
 `videos/observation.images.front/from_timestamp` becomes the
 `"observation.images.front/from_timestamp"` key inside `_episodes.videos`.
@@ -110,7 +106,7 @@ names, including dots such as `observation.state`, remain column names.
 
 | Field | Paimon type | Meaning |
 | --- | --- | --- |
-| `dataset_index` | `BIGINT` | Source ordinal referencing `_datasets.index`. |
+| `dataset_id` | `STRING` | Source UUID referencing `_datasets.id`. |
 | `index` | `BIGINT` | Frame position within the original source, across all its Episodes. |
 | `episode_index` | `BIGINT` | Episode containing this frame, within the same source. |
 | `frame_index` | `BIGINT` | Frame position within the Episode; restarts at 0 for each Episode. |
@@ -119,11 +115,11 @@ names, including dots such as `observation.state`, remain column names.
 | `subtask_index` | `BIGINT`, nullable, when declared | ID resolved through that source's `_datasets.subtasks`; null means no subtask label. |
 | Feature columns | Scalars, `ARRAY`, or `BLOB` | Observations, actions, and other features, mapped as described below. |
 
-Within a table group, `(dataset_index, index)` identifies a frame and
-`(dataset_index, episode_index)` identifies an Episode. Task/subtask IDs are
+Within a table group, `(dataset_id, index)` identifies a frame and
+`(dataset_id, episode_index)` identifies an Episode. Task/subtask IDs are
 also scoped to a source. For example, two sources can both contain frame
 `index=0`, Episode `episode_index=0`, and task `task_index=0` with different
-task texts; `dataset_index` keeps them distinct. The training reader derives
+task texts; `dataset_id` keeps them distinct. The training reader derives
 continuous sample positions at read time without rewriting the stored IDs.
 
 ### Mapping the native LeRobot v3 format
@@ -152,19 +148,6 @@ videos become `BLOB`, so each camera remains a feature column rather than a
 separate Paimon table. See [video import options](#load-lerobot-dataset-v3)
 for optional Episode remuxing.
 
-### Single-source shorthand
-
-For a new import with a single source string and no `dataset_name` or non-empty
-`extra_dataset_fields`, `load_from_lerobot("robot_data", "/data/source_a")`
-still creates the same three tables. Its Dataset row has `index=0` and
-`name=null`, but frames and Episodes do not gain a source `dataset_index`.
-Episode fields such as `data/chunk_index` and `stats/action/mean` retain their
-original names and types instead of being grouped into VARIANT columns.
-
-To use the source-indexed schema above, or to append more sources later, pass
-`source=["/data/source_a"]`. Supplying `dataset_name` or non-empty
-`extra_dataset_fields` also selects this schema, even for one source.
-
 ## Load LeRobot Dataset v3
 
 `load_from_lerobot` imports a local directory, FileIO URI, or Hugging Face
@@ -224,7 +207,8 @@ source is rejected before creating tables to avoid losing original frame IDs.
 Sources must share robot type and compatible definitions for common features. Each source retains its own
 FPS from `info`; different source frame rates are handled automatically. Use
 `PaimonLeRobotDataset` to train from mixed-FPS groups. Pass `dataset_index=N` to
-that reader to select one source. Frame-rate handling is automatic and does not
+that reader to select the source at position `N` in UUID order; this reader
+argument is not the stored UUID. Frame-rate handling is automatic and does not
 require an import option.
 
 ### Schema evolution
@@ -308,8 +292,7 @@ removing suffixes, and names that overwrite standard `_datasets` columns are
 rejected before creating target tables. Training readers project only required
 source metadata and skip extra text columns.
 
-Append another Dataset to a source-local group with `append_to_lerobot`.
-Create the group by importing a list, including a one-element list for one source:
+Append another Dataset to a source-local group with `append_to_lerobot`:
 
 ```python
 conn.load_from_lerobot("robot_merged", ["/data/source_a", "/data/source_b"])