fix(l1): retry failed extraction outputs

[YOMXXX]

Summary

• Treat non-empty, non-JSON L1 extraction responses (for example provider content-filter refusals) as extraction failures instead of empty successful results.
• Propagate extractL1Memories().success === false through the standard L1 runner so the pipeline does not advance the L1 checkpoint cursor after failed extraction.
• Add regression coverage for both the extractor classification and checkpoint-preservation path.

Why

When a provider returns a refusal/content-filter message, the current parser returns an empty extraction. The L1 runner then marks the L0 cursor complete, so the affected messages are skipped permanently and later L2 runs have nothing to process. This matches the silent extracted=0, stored=0 failure mode reported in #104.

Tests

• npm test -- src/core/record/l1-extractor.test.ts src/utils/pipeline-factory.test.ts
• npm test
• npm run build

Closes #104

[Maxwell-Code07]

We have received multiple PRs submitted by you. Thank you for your consistent contributions. We will review them one by one and provide feedback accordingly.