perf(ibd): drop never-written confirm/query/store meters

[rearden-grok]

Summary

X-01 step (1) from the complexity inventory: delete never-incremented confirm/query/store perf meters so ibd: perf / ibd: sizes cannot name a counter that is always zero.

Shipped DEBUG no longer prints recon/wire/resolve, parent_io, miss_p, cold_idx, tip_gc, recent_pub, annotate pread=, spend_mix i=/skip=, pstore, or always-zero heap recent= occupancy. Live lookup / load / scripts / write stage tokens stay (load=, script=, write=, pin/assemble/class_a/spend/tweaks/pins/head_sub/drain_join/dequeue). write= no longer includes always-zero tip_gc/recent_pub. Last-pin / last-write snapshots stay.

Also deletes unused HeadLookupStats sample/snapshot (never consumed by ibd: perf) and the matching store fetch_adds.

Not in this PR: instance-owned stats handle, test mutex / cfg(not(test)) twins, table-driven perf_log (X-01 steps 2–3).

Test plan

• Red then green: format_lines_omit_never_written_inventory_tokens drives shipped format_info / format_debug / format_sizes.
• Targeted cargo test -p rbitcoin-net --lib ibd::perf_log, plus query sampler_stats / size_snapshot_counts, consensus last_write_phase_stats, store segmented_head.

[rearden-grok]

Review: X-01 step 1 against complexity-inventory-2026-09-07.md

Verdict: approve once required checks are green, with one should-fix (tip_gc half-delete) that I would fold into this PR rather than a follow-up. The App token is the PR author so this is posted as COMMENT.

What I verified

• Every deleted static is dead. 40 static … Atomic* definitions removed. On master none has a production fetch_add/store outside #[cfg(test)], except the S-02 (LOOKUP_*/ROLLS/SEALS in segmented_head.rs) and Q-01 (PSTORE_*) groups, which are written but had no consumer (sample_head_lookup_stats/snapshot_head_lookup_stats had zero call sites outside the store crate; note(...,0,0,0) was the only pstore writer). Matches the inventory's X-01 / I-05 / C-01 / Q-01 / Q-02 / S-02 lists exactly.
• No live token dropped. Every removed format token (recon_ms/us, wire_*, resolve_*, parent_io, miss_p, cold_idx, cold_dec, edges same=, spend_mix i=/skip=, pread=, recent_pub, tip_gc in write=, creates=, phases hdr=/dec=/put=/new=, pstore, heap recent=) maps to a deleted counter or a hard-coded zero. load=/script=/write= and the pin/assemble/class_a/spend/tweaks/pins/head_sub/drain_join/dequeue inventory are intact; WriteStageSample::stage_ms/ns inventory shrinks only by the two always-zero terms.
• Remaining confirm_load_stats / confirm_phase_stats / archive_phase_stats statics are all live — spot-checked ~60 of them; each has a production writer (pin.rs, lookup.rs, phases.rs, block/mod.rs, connect.rs, reconstruct.rs, or a note_* fn). wf_body_store / sh_collect_pin from I-05 are live (reconstruct.rs:14, connect.rs:647) so keeping them is correct.
• Red → Green is real. format_lines_omit_never_written_inventory_tokens (7398bd5) fails on master for the unconditional tokens (tip_gc=, recent_pub=, parent_io=, cold_idx=, pread=, creates=, pstore, recent=, dbg edges same=) and drives shipped format_info / format_debug / format_sizes. Several of the other assertions (recon_ms, wire_ms, resolve_ms, miss_p, spend_mix) were append_nz/conditional on master and so were already green with a zero sample — harmless, just not load-bearing.
• Deleting lookup_stats_sample_and_snapshot_surface is right; it asserted nothing (let _ = any;).
• CHANGELOG / OPERATOR.md / docs/ibd-memory.md updated for the DEBUG token shrink as the inventory asks.
• Second force-push (6c684c14 → 98bba3d7) is a same-tree, same-message amend (CI retrigger) — fine per AGENTS.

Should fix (in this PR)

tip_gc is half-deleted. The write= token is gone, but the always-zero meter survives end-to-end:

• confirm_run/phases.rs:352 let tip_gc_ns = 0u64; returned from post_commit as the second tuple element,
• threaded through write.rs:240/258/294 into LastWritePhases.tip_gc_ns,
• stored in LAST_WRITE_TIP_GC_NS (consensus/lib.rs:210/253/269),
• and printed as tip_gc={}ms in the ibd: confirm write slow … INFO line (ibd/confirm/mod.rs:1465-1475).

The PR's stated contract is that no log line names a counter that is always zero, and the CHANGELOG entry says tip_gc is dropped — the INFO line still prints it. Delete the plumbing: post_commit returns just spend_ann_ns, drop the LastWritePhases field / static / snapshot field, and the tip_gc= token from the slow-batch line. note_last_write is called from write.rs:283 and the coverage test, so the fix is small and the existing last_write_phase_stats test pins it.

Non-blocking

• Q-03 trimmed instead of deleted. ConfirmLoadStats + #[cfg(test)] confirm_load_stats::note are test-only scaffolding (the only writer of that struct is query_tests::sampler_stats; every static it feeds already has a real writer in confirm_run/pin.rs). The PR removes 12 of its fields, which is wasted motion on a struct the same inventory step says to delete. Suggest: delete the struct, note, and the pub use; have sampler_stats assert the sampler via sample_and_reset after driving the real accrual (or drop that block). Then confirm_load.rs is two Query passthroughs and can fold into confirm_parent_cache.rs as Q-03 says. Fine as a follow-up if you want this PR to stay pure deletion.
• Inventory suggested order was X-04 (visibility pass) before this step so rustc lists the truly dead items. Skipping it means completeness here rests on grep; I did not find any additional never-written meter in the modules touched, but the X-04 pass will be the real check.
• sample_assemble_prevout_detail_and_reset now returns a 4-tuple of counts; the doc comment is updated, name is fine.

No behavior change outside DEBUG/INFO log text; no SCHEMA impact.

[rearden-grok]

tip_gc is only dropped from ibd: perf write=. It is still printed as tip_gc={}ms in the ibd: confirm write slow … INFO line (crates/rbitcoin-net/src/ibd/confirm/mod.rs:1465-1475), fed by post_commit's hard-coded let tip_gc_ns = 0u64; (confirm_run/phases.rs:352) through LastWritePhases.tip_gc_ns / LAST_WRITE_TIP_GC_NS. Delete that plumbing in this PR so the entry is true.

[rearden-grok]

Q-03 in the inventory: this struct and confirm_load_stats::note are #[cfg(test)]-only scaffolding — the only writer is query_tests::sampler_stats, and every static note feeds is already accrued in production by confirm_run/pin.rs. Trimming 12 fields off a dead struct is wasted motion; delete the struct + note + pub use (and fold the two remaining Query passthroughs into confirm_parent_cache.rs). OK as a follow-up if you want this PR to stay pure meter deletion.