fix(ask): raise output budget to 8192 and fail on empty strategy

[lfnovo]

Summary

The three Ask stages (search strategy, per-search answers, final synthesis) in open_notebook/graphs/ask.py were hardcoded to max_tokens=2000, well below the 8192 that chat and transformations already use. Two failure modes followed from that, both returning HTTP 200:

• Token-dense languages (Japanese, etc.) hit the cap and the answer was cut mid-sentence.
• Reasoning models (Qwen3 on Ollama, thinking-enabled Claude/GPT) spent the whole budget thinking. After clean_thinking_content() the strategy had blank search terms, vector search returned nothing, and Ask answered "no documents found" from a corpus that contained the answer.

This PR:

• Shares a single ASK_MAX_TOKENS = 8192 budget across the three stages, aligned with graphs/chat.py and graphs/transformation.py. No new environment variable: per-model tunables are being consolidated under [Design] Model & Provider Configuration (advanced params, presets, health) #947 / feat(chat): expose temperature/top_p controls #830, so this keeps the fix to a sane default in the meantime (same reasoning as the parking note on fix: make Ask output token budget configurable #1247).
• Drops blank search terms from the strategy and raises ExternalServiceError with an actionable message when none remain, instead of running empty searches.
• Skips thinking-only partial answers so they don't pollute the final synthesis.

Closes #1221

Test plan

• uv run pytest tests/ — 679 passed (9 new tests in tests/test_ask_graph.py cover the budget on all three stages, blank-term filtering, the empty-strategy error and the empty partial-answer skip).
• uv run ruff check . — clean.
• uv run python -m mypy . — clean.

[cubic-dev-ai]

All reported issues were addressed across 3 files

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[lfnovo]

Addressed both findings: parse_thinking_content now treats an unclosed <think> block as reasoning-only (so a truncated partial answer is dropped rather than leaked), and the empty-strategy error now points at the Ask page's advanced model options. Tests added for both.

@cubic-dev-ai review please

[cubic-dev-ai]

Addressed both findings: parse_thinking_content now treats an unclosed <think> block as reasoning-only (so a truncated partial answer is dropped rather than leaked), and the empty-strategy error now points at the Ask page's advanced model options. Tests added for both.

@cubic-dev-ai review please

@lfnovo I have started the AI code review. It will take a few minutes to complete.

[cubic-dev-ai]

No issues found across 5 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

_{Re-trigger cubic}