perf(sessions): skip unchanged host bookkeeping writes on idle passes

[devin-ai-integration]

Summary

• An idle history pass no longer commits anything to the session stores. Before this change, each pass rewrote every provider's host-coverage://<provider>/v1 row and the Kimi and Pi discovery frontiers, even when nothing had changed. That was 9 WAL commits per idle project pass, plus about the same on user-sessions.db. The daemon runs this pass every minute (HISTORY_IDLE_RECHECK_INTERVAL).
• The new revise_host_record (runtime/source.rs) writes a record only when its byte_offset or file_id changes. Those are the only fields readers check. mtime is still bumped on a real change, which lets a record move byte_offset backwards past the monotonic cursor guard.

Motivation

Refs #2521, Refs #2658.

I re-measured after #2777 and #2948. The capture path already reads only the objects a commit names. In load_current_mutations, SourceObjectCoverageV1::Complete is reached only for complete-snapshot-absence sources. Next I looked at an idle daemon with nothing being ingested, using an isolated profile and project. Temporary SQL probes, since removed, plus parse_offsets snapshots showed what changed each minute: host-coverage://{claude,cursor,kimi,opencode,pi}/v1 and host-frontier://{kimi,pi}/discovery/v1 were rewritten on both stores with only mtime incremented. Every such commit adds WAL frames that readers and checkpoints then reread for as long as the daemon is idle. #2658 lists this idle traffic as the thing that makes whole-daemon bytes per message unmeasurable.

Changes

• crates/tracedecay-sessions/src/runtime/source.rs: persist_host_provider_coverage now goes through the new revise_host_record, which skips unchanged records.
• hosts/kimi.rs, hosts/pi.rs: the discovery frontier writes use revise_host_record. Reaching the end on an already-reset frontier no longer writes.
• ingest/project_provider.rs, ingest/user_provider.rs: removed the Codex callers' stored_coverage != Some(coverage) guard. The shared write now owns that check for every provider.
• crates/tracedecay-session-runtime/tests/session_store_read_cost.rs: new idle_history_pass_commits_nothing. It runs the production project ingest pass twice over an empty host home on a registered store. It asserts that the first pass commits and the second appends no WAL commit frames.

Test plan

• bash scripts/require-exact-test.sh cargo test -p tracedecay-session-runtime --features test-helpers --test session_store_read_cost -- idle_history_pass_commits_nothing --exact: 1 passed. Without the tracedecay-sessions change it fails with left: Some(9) right: Some(0).
• cargo test -p tracedecay-sessions --lib -- runtime::hosts runtime::ingest runtime::source: 353 passed.
• cargo clippy -p tracedecay-sessions -p tracedecay-session-runtime --all-targets --features tracedecay-session-runtime/test-helpers -- -D warnings: clean. cargo fmt --all -- --check: clean.
• Tested manually on a real daemon built from this branch, with an isolated HOME/TRACEDECAY_DATA_DIR and an enrolled project. Before the change, coverage and frontier rows were rewritten every idle minute. After it, parse_offsets and both stores' WAL mxFrame were byte-identical across 130 s (two idle passes).

Not covered: operator-scale read volume (the 2 TB / 3 h in #2521) is not re-measured here. The per-message store-growth target in #2658 is unchanged by this PR.

Checklist

• CHANGELOG.md: not updated; release-please generates it from commits
• No secrets, credentials, or .env files included
• Breaking changes documented (none)

Link to Devin session: https://app.devin.ai/sessions/868821e98d0743dc83648344702f3479
Open in Devin Desktop: https://app.devin.ai/desktop/session/868821e98d0743dc83648344702f3479?variant=devin
Requested by: @ScriptedAlchemy

---

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring

[changeset-bot]

⚠️ No Changeset found

Latest commit: 56adcc9

Merging this PR will not cause a version bump for any packages. If these changes should not result in a new version, you're good to go. If these changes should result in a version bump, you need to add a changeset.

Click here to learn what changesets are, and how to add one.

Click here if you're a maintainer who wants to add a changeset to this PR

[devin-ai-integration]

🔍 Devin Review: 1 flag

Not posted on this PR by your GitHub settings — view it in Devin Review. (Configure)