[nullPointerEnjoyer]

Summary

Keeps the peer's send stream continuous during the block download instead of stopping between batches:

• Block-batch top-up: while blocks are in flight, the next portion of the staged headers is requested as soon as the outstanding count drops below half of max_request_blocks_count, so the peer starts serving the next batch without waiting for the current one to fully drain. Total outstanding requests stay within max_request_blocks_count, which is also the serving limit peers enforce on their side.
• Header look-ahead window: header requests keep up to 2 × msg_header_count_limit headers staged ahead of the download, so header fetch overlaps block fetch. Requests are issued only while the window is below the target and the previous response actually staged new headers, so repeated or purely overlapping lists cannot turn into request loops (the consecutive known-full-lists backpressure is unchanged).
• Overlap trimming: header responses received during a download are anchored at the last processed block, so they legitimately repeat in-flight/staged headers. The repeated prefix is skipped; the staged window stays a single connected chain onto a known or in-flight header (branching lists keep the window connected up to the branch point; wholesale divergences replace it, as before).
• Known-block filtering: staged headers whose blocks arrived from another peer in the meantime are filtered out before mid-flight requests, mirroring the existing batch drain path.

Out-of-order and unsolicited block handling, stall detection, the fork download budget, and wait_for_clock_diff are unchanged.

Verification

• cargo test -p p2p --lib: 340 passed (330 pre-existing + 10 new/extended), covering: pipelined request continuity, no duplicate/out-of-limit requests, overlap dedup without spin, window bound under divergent zero-overlap lists (capacity cap exercised at exactly the target), reorg-mid-download connectivity, known-lists grace after a download, budget deferral + retry during a download, and lockstep degradation at small request limits.
• ./do_checks.sh (fmt / deny / vet / clippy ×2 / codecheck / wasm-docs): pass.
• cargo test -p chainstate-test-suite: pass.

Notes

• No consensus surface; p2p-local change against master post-perf(experimental): IBD and indexer performance improvements  #2156.
• Measured mainnet A/B numbers: pending (existing ibd-ab harness, run D vs run C baseline 5,164 s). Expected effect is largest on high-RTT peers; will be appended when available.
• Draft / held from merge until the 1.5.0 content set is final.