perf(state): measure contextual verification phases

[evan-forbes]

Summary

• measure the full contextual interval, initial checks, parent-chain work, new
  chain construction, and state-writer snapshot cloning;
• separate the cumulative UTXO snapshot from transparent spend checks;
• measure shielded anchors, contextual block construction, and the parallel
  state update;
• separate the parallel chain clone from Chain::push;
• measure header-transition preparation and commit on success and error paths;
• emit matching mined-only series for every phase on the mined-block path.

Motivation

The preserved memo-enabled field run reduced the median transaction phase to
1.795 ms. Its median state commit remained 34.338 ms. PR #781 separates writer
queue delay from writer execution. This PR splits writer execution so the fleet
can identify the next material optimization without inferring phase costs from
aggregate quantiles.

Safety

The change adds elapsed-time capture and histograms only. It does not move a
check, change state ownership, or alter chain selection.

The outer parallel_update duration includes task scheduling and completion.
The parallel_task.* durations overlap because block commitment, Sprout anchor
validation, and chain update run concurrently. Only parallel_update measures
their critical-path wall time.

Benchmark scope

This PR adds observability. It does not claim a runtime speedup. A synthetic
microbenchmark would not show whether these histograms partition the production
contextual interval. The state tests exercise every instrumented success and
error path. The next fleet run must compare each mined-only phase count with
the outer contextual count, then rank phase means from matching campaign
deltas. Concurrent parallel_task.* means must not be added together.

Verification

• cargo test -p zakura-consensus -p zakura-state --lib --locked
  • 210 consensus tests passed
  • 520 state tests passed and 3 tests remained ignored
• cargo clippy -p zakura-state -p zakura-consensus --all-targets -- -D warnings
• cargo check -p zakura-state -p zakura-consensus --all-targets --locked
• cargo fmt --all -- --check
• git diff --check
• ./scripts/changelog.py check

This PR adds docs/changelog/unreleased/782.md with an internal-only marker.

Stack

• Base and parent: perf(state): measure mined block admission timing #781.
• Children: none.
• This branch contains feat(mining): advertise admitted blocks optimistically #748 as its root dependency and perf(state): measure mined block admission timing #781 as its parent.
• This branch does not contain feat(mining): submit prepared blocks by solved header #754, perf(consensus): accelerate block verification #776, perf(state): avoid contextual UTXO copies #779, perf(state): share contextual snapshot maps #780, or feat(consensus): relay expected-work blocks before contextual commit #773.

[czarcas7ic]

The copied unspent outputs now stay in memory for the rest of validation, so we use more memory than before. We could free that copy as soon as the spending checks finish, like the old code did.