Fix log decompression bounds and CCM time filters

[cursor]

Description

Fixes two critical correctness issues found in recent changes:

• ZIP log blobs with unknown uncompressed sizes are now bounded by the configured decompressed log limit, preventing compressed log payloads from inflating beyond the budget.
• CCM time filters that were advertised but not implemented (THIS_QUARTER, THIS_YEAR, LAST_QUARTER, LAST_YEAR, LAST_3_MONTHS, LAST_6_MONTHS) now map to explicit ranges instead of silently falling back to a last-30-days query.

Type of Change

• Bug fix
• New feature
• Refactor
• Documentation
• Other

Checklist

• Tests pass
• Typecheck passes

Validation:

• pnpm test tests/utils/log-resolver.test.ts tests/registry/registry.test.ts
• pnpm test
• pnpm typecheck
• pnpm build
• git diff --check

  

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[cursor]

No findings.

Assumptions / Open Questions

I assumed THIS_YEAR and THIS_QUARTER are intended to be period-to-date filters, because the implementation and added regression note/test explicitly encode that behavior. If CCM expects full-period semantics analogous to the existing THIS_MONTH branch, that contract should be clarified separately.

Overall Assessment

This PR aligns with Sunil's architecture standards overall. The fixes stay narrowly scoped to the affected helpers in src/registry/toolsets/ccm.ts and src/utils/log-resolver.ts, keep the logic in the right layer, harden the log path to fail loudly on oversized decompression, and add focused regression coverage without broad churn. The tasks/todo.md update is appropriate for this repository's required task-tracking workflow and matches the change set.

Verification / Residual Risk

Verified locally with:

• pnpm test
• pnpm typecheck
• pnpm build

Residual risk is low. The main remaining gap is that the CCM regression test proves THIS_YEAR, but the other newly added date branches (THIS_QUARTER, LAST_QUARTER, LAST_YEAR, LAST_3_MONTHS, LAST_6_MONTHS) still rely on untested date math. I did not find an active defect in those branches, but they are the remaining regression surface for this PR.

  

_{Sent by Cursor Automation: Sunil On Demand Architecture Review}