fix: port high-value upstream stability fixes (wave 1)

[sakitam-fdd]

Scope

This is Wave 1 of the reviewed upstream stability ports.

Tencent#2763 — batch reparse uses current parser rules

• Clears only upload-time parser-engine rule snapshots for nil-config batch reparses.
• Preserves document-specific chunking, multimodal settings, and parser options.
• Prevents failed documents from silently reusing obsolete parser selection after the KB parser configuration changes.

Tencent#2711 — harden LLM rerank fallback for reasoning models

• Disables thinking for deterministic rerank scoring and reserves output tokens.
• Requires exactly one score per passage instead of padding/truncating incomplete responses.
• Stops later rerank batches after unusable output and falls back to original retrieval scores.
• Uses the upstream final test stub fix that returns a closed stream channel.

Porting notes

• Base: main@4e25684b8ff55a70a55c03730d81457c14521d3c.
• fix(knowledge): use current parser rules for batch reparse Tencent/WeKnora#2763 was applied as a focused atomic commit.
• fix: harden LLM rerank fallback for reasoning models Tencent/WeKnora#2711 diverged from current main, so it was applied through GitHub's three-way merge onto the Wave 1 branch to preserve intervening knowledge_search.go changes.
• Final diff against main is limited to the expected four files.

Validation

Upstream focused validation reported:

• fix(knowledge): use current parser rules for batch reparse Tencent/WeKnora#2763: focused service tests, go vet, and diff-scoped lint passed.
• fix: harden LLM rerank fallback for reasoning models Tencent/WeKnora#2711: focused tests, race tests, go vet, and git diff --check passed.

The execution environment used for this port cannot perform an outbound-DNS-backed full checkout, so this fork's GitHub Actions checks are the authoritative target-branch validation.

No public API or schema changes. No intended breaking changes.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 8838c562-4b5e-4936-880c-af60850965d5

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}