fix(memory): resample unthrottled after allocator trim

[devin-ai-integration]

Summary

• After an allocator trim, the trim reclaimer now takes a fresh, unthrottled memory reading and publishes it. It no longer goes through the throttled checkpoint sampler, which could reuse a stale reading.
• This fixes the Devin Review / Codex P1 finding from fix(memory): resample pressure right after allocator trim #2596. That PR, and fix(memory): resample unthrottled after allocator trim #2856, both closed without this change landing. On master the trim callback still calls sample_for_checkpoint().

// register_process_allocator_pressure_reclaimer_v1
- pressure.sample_for_checkpoint();      // throttled (~10ms); may return the standing OverBudget state
+ pressure.resample_after_reclaim();     // canonical sampler, unthrottled, no reclaimer re-entry

fn resample_after_reclaim(&self) -> Option<ResidentMemoryPressureStateV1> {
    let sample = (self.sampler)()?;
    self.publish_observation(sample.admission_bytes());
    self.publish_over_budget_gauge();
    Some(self.state())
}

Motivation

The pressure latch only clears once memory drops below the low watermark. Suppose an indexing checkpoint sampled within the last RESIDENT_MEMORY_CHECKPOINT_SAMPLE_INTERVAL_V1. The post-trim call then returns the standing state without reading memory again. run_pressure_reclaimers doesn't resample afterwards either, so the latch stays OverBudget even though memory has recovered. Admissions keep being refused until the maintenance sampler next runs.

Changes

• crates/tracedecay-runtime-core/src/resident_memory.rs: adds a private resample_after_reclaim and calls it from the allocator trim reclaimer. It publishes the cgroup-aware admission bytes and the gauge. It does not call run_pressure_reclaimers, so the reclaimers can't re-enter.
• crates/tracedecay-runtime-core/src/resident_memory/tests.rs: adds allocator_trim_readmits_growth_when_memory_recovers_inside_checkpoint_throttle. The test checks the behavior through ProcessResidentMemoryV1::reserve, the API admission callers actually use, with literal inputs:
  1. Memory is at the 1 GiB limit. A checkpoint sample starts the throttle window, and a 16 MiB reserve is refused with ObservedOverBudget.
  2. Memory drops to 64 MiB and the pressure pass runs.
  3. The same reserve now succeeds, and the ledger records 16 MiB used.

Test plan

• cargo nextest run --workspace --no-fail-fast passes (left to CI)
• cargo clippy -p tracedecay-runtime-core --all-targets -- -D warnings clean
• cargo test -p tracedecay-runtime-core --lib resident_memory: 52 passed. With the old sample_for_checkpoint() call restored, the new test fails at the post-recovery reserve, which is still refused.
• cargo fmt --all -- --check

Checklist

• CHANGELOG.md updated (under [Unreleased] if no version bump)
• No secrets, credentials, or .env files included
• Breaking changes documented (if any): none

Link to Devin session: https://app.devin.ai/sessions/9f41edacea9a4fffb64bd90be9914a12
Open in Devin Desktop: https://app.devin.ai/desktop/session/9f41edacea9a4fffb64bd90be9914a12?variant=devin
Requested by: @ScriptedAlchemy

---

Local verification (land-open-prs)

On the head rebased onto master 9d24792:

• Fails before: with the reclaimer reverted to pressure.sample_for_checkpoint(), resident_memory::tests::allocator_trim_readmits_growth_when_memory_recovers_inside_checkpoint_throttle fails 10/10 (tests.rs:1202, growth refused).
• Passes after: cargo test -p tracedecay-runtime-core --lib -- resident_memory --test-threads=16: 52 passed.
• Runtime journey: cargo test -p tracedecay-cli --features alloc-mimalloc --test refresh_memory_journey (shipped daemon cold-indexing and refreshing a 300-file heavy corpus under a 6 GiB limit): 2 passed in 149 s.
• cargo clippy -p tracedecay-runtime-core --all-targets -- -D warnings clean; cargo fmt --all -- --check clean. No cfg-gated code touched.

This carries the remaining part of #2596 (closed); #2851 landed its first two commits.

[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: 5b65159

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: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.