docs: clock inventory for #821 step 1 (100 sites, 886463ef)

[Hero-Gamer]

Step 1 of #821 - inventory only, no prod change.

HEAD: 886463e Merge #823 - latest master Sep 27, includes #822/#813/#773/#769/#746
Command: rg -n "SystemTime::now()" crates/ --type rust -g '!tests' = 100 sites
Verified: rg | cut -d: -f1,2 vs docs table = 100/100 match, comm -23 empty, comm -13 empty after fix 923->928, cargo fmt --all --check=0, no trailing WS, 120 lines total

Breakdown (100 rows):

• 35 MIGRATE (consensus header.rs/lib.rs/params.rs, net peers.rs:1249 3 sites, service.rs 3, peer.rs 4, seeds.rs 3, tx_relay.rs:928 manual mock, mempool accept.rs/orphanage.rs, rpc mine.rs etc)
• 65 LEAVE (store bulk_io 10 bdz 5 spender_table etc use unwrap_or(0) not unwrap, log api_log, query, esplora, cli, tmp uniqueness tx_relay.rs:3753)
• clock.rs:44 root wall source = LEAVE (impl NodeClock::now() - the atomic clock)

Adversarial findings for Step 2/3:

• peers.rs:1249 mock_now AtomicU64 Acquire/Release + set_mock_now() bypasses hub.clock
• tx_relay.rs:625 mock_now AtomicU64 Relaxed + note_mock_now() bypasses hub.clock
• chain.rs: hub.clock.set_mock() third clock = split-brain
• tx_relay.rs:925 if mock!=0 tape + 928 SystemTime::now() = 0-sentinel prevents mock to epoch 0, Relaxed vs Acquire race on ARM (M4), INV freeze risk if mock set for chain but not relay (mempool bloat / peer timeout bypass hides eclipse test)

All LEAVE marked log/perf per OPERATOR.md #688 item 2.1.

Unblocks Step 2 PeerManager NodeClock injection and Step 3 TxRelay shared NodeClock.

Ref #740 unified-clock-api, part of #821, tracks #688 item 2.1

[rearden-grok]

Closing as superseded by #837. #837 already includes docs/clock-inventory.md along with the Step 2 implementation, so the inventory can land with that PR and does not need a separate pull request. Please keep the inventory there and continue review on #837.