Suppress recovered rapid heap growth warnings

[kojiwakayama]

Summary

• Delay Rapid heap growth detected by one memory-monitoring interval and only warn if heap remains more than 100MB above the pre-spike baseline.
• Preserve the existing warning message for sustained growth, while adding observedGrowthMB to distinguish the original spike from the sustained delta.
• Bump deno.json and src/utils/version-constant.ts from 0.1.817 to 0.1.818.

Runtime evidence addressed

The production sample from #3200 showed a cold render/cache warmup spike from 99.16MB to 212.69MB, then recovery to 153.01MB on the next sample. This change suppresses that recovered one-interval spike but still alerts when the next sample remains over the 100MB sustained-growth threshold.

Test plan

• Red/green regression: deno test --no-check --allow-all src/utils/memory/profiler.test.ts
• deno test --no-check --allow-all src/utils/version.test.ts src/utils/logger/logger.test.ts src/utils/memory/profiler.test.ts
• DENO_NO_PACKAGE_JSON=1 deno lint src/utils/memory/profiler.ts src/utils/memory/profiler.test.ts
• deno check src/utils/memory/profiler.ts src/utils/memory/profiler.test.ts
• deno task typecheck
• deno task test:unit passed: 2167 tests, 18661 steps.
• git push pre-push checks passed: formatting, lint, typecheck, unit tests.

Known repo gate

• deno task verify:quick currently stops at pre-existing lint:cli-boundary violations in cli/shared/server-startup.ts, unrelated to this diff.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 49930cc7d1

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[kojiwakayama]

Addressed the inline feedback about restart baselines.

startMemoryMonitoring now seeds the rapid-growth baseline from the current heap via getInitialRapidHeapGrowthState(getHeapStats().usedHeapSizeMB) instead of resetting lastHeapUsed to zero. Added a regression test for the stable warm-heap restart case so a 153MB heap that remains flat does not create a pending spike or sustained warning.

Verification after the update:

• deno test --no-check --allow-all src/utils/memory/profiler.test.ts
• DENO_NO_PACKAGE_JSON=1 deno lint src/utils/memory/profiler.ts src/utils/memory/profiler.test.ts
• deno check src/utils/memory/profiler.ts src/utils/memory/profiler.test.ts
• deno task typecheck
• pre-push checks passed: formatting, lint, typecheck, unit tests (2167 passed, 18662 steps).