feat: add activity processor tag extraction thresholds

[JimmyPettersson85]

Ticket

• https://linear.app/stream/issue/FEEDS-1792/make-activity-processor-tag-extraction-thresholds-configurable

Summary

Regenerates src/gen from the chat OpenAPI spec to pick up two new optional fields on ActivityProcessorConfig:

min_text_length?: number;
min_word_count?: number;

They let a feed group gate text_interest_tags extraction on how much text an activity actually has. Today there is no gate at all — a one-character activity triggers a full LLM call and whatever topics the model invents land in interest_tags, which is what a customer reported as junk tags. Both fields default to 0 (disabled), so existing behaviour is unchanged and the knob is opt-in.

Notes for review

Branch name matches the chat branch on purpose. tests/qa/Makefile resolves VIDEO_JS_SERVER_SDK_BRANCH by looking for a stream-node branch with the same name as the chat branch, and swaps @stream-io/node-sdk to it for the QA run. That is how GetStream/chat#16074's QA tests exercise min_text_length / min_word_count before this SDK is released. (This PR supersedes #338, which sat on a non-matching branch name.)

The regeneration diff is only those 10 lines — src/gen was already in sync with master, so nothing unrelated came along.

Both fields are optional, so this is additive and backwards compatible: existing callers constructing an ActivityProcessorConfig keep compiling untouched.

Verification

• yarn generate:open-api — regenerated, diff scoped to ActivityProcessorConfig
• yarn build — clean
• yarn lint — clean, no new warnings
• yarn test — 18 failed / 3 passed / 3 skipped, identical to main. Every failure is secretOrPrivateKey must have a value from missing STREAM_API_SECRET locally; verified by stashing the change and re-running. A type-only addition cannot affect these.

🤖 Generated with Claude Code