feat(chunk): add chunk get peer diagnostics

[mickvandijke]

Summary

• add ant chunk get --all-peers to query every selected closest peer and print ranked per-peer results
• print XOR distance as a decimal number
• add --peer-count / --peers to choose how many closest peers to try, e.g. 20 instead of the default close-group size

Verification

• cargo fmt --all -- --check
• cargo test -p ant-cli xor_distance_decimal
• cargo clippy -p ant-core --lib -- -D warnings
• cargo clippy -p ant-cli -- -D warnings

Note

• cargo clippy --all-targets --all-features -- -D warnings currently fails on the RC branch outside this diff due multiple saorsa_core versions causing MultiAddr / P2PNode type mismatches in devnet/test-support code.

[grumbach]

A few real issues worth addressing before merge:

• high ant-core/Cargo.toml:40 — swaps ant-protocol from a registry pin to a git branch (rc-2026.5.4), which transitively pulls a second saorsa-core 0.24.5-rc.1 into Cargo.lock. The PR's own description admits cargo clippy --all-targets fails on the RC branch due to multiple saorsa-core versions and MultiAddr/P2PNode type mismatches. Best to land the protocol bump in its own PR, or block this one until the RC merges to crates.io. We don't want the diagnostic PR carrying the dep-bump fallout.

• high ant-cli/src/commands/data/chunk.rs:34 — --peer-count works without --all-peers: it routes the default chunk get through chunk_get_from_closest_peers(addr, N), expanding the queried set well beyond close_group_size. This silently breaks the close-group invariant assumed by AIMD/budget accounting and the cache-write path — which could be contributing to the exact symptom you're chasing. Gate --peer-count behind --all-peers, or cap N <= close_group_size in the early-return path.

• high ant-core/src/data/client/chunk.rs:362 — chunk_get_from_closest_peer_group dispatches all N peer GETs concurrently via FuturesUnordered with no per-command deadline and no concurrency cap. With --peer-count 200 this fans out 200 simultaneous chunk RPCs and waits for the slowest 10s timeout. Bound with buffer_unordered(K) and an overall timeout, or document the diagnostic-only intent in --help.

• medium ant-cli/src/commands/data/chunk.rs:54-72 — CLI claims success ("Saved closest successful chunk to ...") whenever any peer returned the chunk, even if N-1 timed out. The summary line is honest but the headline isn't. Suggest Saved chunk (1/N peers responded successfully) so the operator can't misread a near-total failure as a clean hit.

• medium (tests) — only xor_distance_decimal has new tests. chunk_get_from_closest_peer_group (ordering, summary counts, cache-write on partial success) and the CLI flag plumbing have none. At least a unit test that exercises the result-sorting and the summary tallies with stubbed chunk_get_from_peer outcomes.

• low ant-cli/src/commands/data/chunk.rs:166,168,186,203,211,215,219 — format args inconsistently inlined. Project style is format!("foo {var}"); bind locals (let display = path.display();) where needed.

• low ant-core/src/data/client/chunk.rs:368 — pre-dispatch peers.sort_by_key(...) is dead work; results are re-sorted by xor_distance at line 401 after FuturesUnordered completion order randomises them. Drop the pre-sort.

No unwrap/expect/panic introduced in non-test paths. peer_xor_distance consolidation into mod.rs and the quote.rs call-site updates are a clean dedup.

[dirvine]

Review: PR #97 — feat(chunk): add chunk get peer diagnostics

I reviewed the current diff (4 commits merged, latest 3191af9). Mick has addressed all of Anselme's review items from the previous round. Checking each:

Previously flagged items — status

#	Issue	Status
1	Cargo.toml ant-protocol git dep pin	✅ Removed — no Cargo.toml diff present anymore
2	--peer-count works without --all-peers	✅ Fixed — requires = "all_peers" in clap + runtime bail
3	Unbounded concurrent fan-out in diagnostic sweep	✅ Fixed — buffer_unordered(concurrency_limit) + diagnostic_peer_get_overall_timeout
4	Misleading success headline on partial failure	✅ Fixed — shows "x / y peers responded successfully"
5	Insufficient test coverage	✅ Added — sorting, timeout scaling, summary tallies, CLI flag validation
6	Inconsistent format-arg style	✅ Fixed — consistent {var} inlining with let display bindings
7	Dead pre-sort before FutUnordered	✅ Removed — sort_chunk_peer_get_results runs after collection

Additional observations

1. chunk_get now routes through chunk_get_from_closest_peers — This changes the hot-path to query close_group_size peers and return the first success, matching the existing behavior of the old inlined loop. No regression risk.

2. chunk_get_from_closest_peer_group caches on partial success — Caching found chunks during diagnostics is harmless and improves cache-hit rates on retries. Fine.

3. Schema-safe — No new dependencies, no protobuf/serialization changes, no public API additions beyond the CLI.

4. closest_peers(count) extracted from close_group_peers() — Clean dedup. The quote.rs xor_distance → peer_xor_distance migration is equally clean.

Verdict

Approve. All review-blocking items resolved. Clean, well-tested, and safe for the RC branch.