fix(observability): pin one telemetry snapshot per Logs and Trace response #280

The whole-PR review of #272 identified a low-severity, pre-existing response consistency gap outside its five inline findings: #272 (review) .

Logs queries entries and histogram separately. Trace queries a candidate, reads its index, and fetches related logs in separate snapshots. A telemetry publication between those operations can make sections of the same response describe different active file sets.

Pin one bounded telemetry snapshot for every statement/index read that contributes to a single Logs or Trace response. Read cache acknowledgements and contributions transactionally under that snapshot; preserve immediately visible uncached files, exact clipped windows, redaction, cancellation and publisher progress. Define transaction/lock ownership once rather than adding retries or snapshot fallbacks.

Acceptance: deterministic tests publish, compact or retain telemetry between response stages and prove entries/counts and trace/log sections describe one file set. Cancellation releases all resources; publishers still progress under the existing reader grace period; the full native and race suites pass.