fix(generation): stop creative-space generations failing on Anki metadata validation

[mrsibe]

What does this PR do?

Makes the Anki, mind map and quiz generators stop discarding an entire result when the model omits a field the prompts never required. Concretely, Anki card generation now succeeds instead of always ending in a failed item, and chunk IDs are treated as the best-effort field the prompts describe.

Why?

In the creative space, creating Anki cards always failed after the model finished streaming, and chart/quiz creation was exposed to the same class of failure. The root cause is that the Zod schemas used for streamObject are stricter than the prompts:

• Anki (reproducible, always fails): createAnkiSchema required metadata.cardTypes: string[], but the Anki prompt's output format only mentions metadata.totalCards. The model follows the prompt, omits cardTypes, and final validation throws AI_NoObjectGeneratedError: No object generated: response did not match schema with path: ["metadata","cardTypes"], expected "array", received undefined.
• chunkIds (all three): aggregateNotebookContent never passes chunk IDs to the model — only the chunk text — while the prompts ask for chunk IDs only "whenever possible" (尽可能). Yet chunkIds was required, so one omitted node/card/question field failed the whole generation.
• Anki text caps: the schema imposed front max(300), back max(500), etc., which the Anki prompt never states, so a long answer could fail the same way.

Related issue

None.

What changed?

• AnkiCardService: metadata is now optional and totalCards/cardTypes are derived from the generated cards; chunkIds defaults to []; the undeclared max() caps on card text are removed.
• MindMapService / QuizService: chunkIds defaults to []. The existing validateAndClean* code already filters invalid IDs and rebuilds the chunk mapping, so this only affects the "model omitted them" case.

Verification

• npm run typecheck — pass
• npm test — 49/49 pass
• npm run build — pass
• Manual (Linux, electron-vite dev): ran the real generation path against the configured openai-completions / api.deepseek.com connection.
  • Before: Anki failed with No object generated: response did not match schema (missing metadata.cardTypes).
  • After: Anki returned 4 cards with cardTypes=basic; mind map and quiz already succeeded and still do.

No new tests were added: the generation services are not unit-testable without a model connection, and the schemas are module-private. The prompt/schema mismatch is covered by the manual verification above.