Chore/model bump

[GilesStrong]

Adds reasoning levels to models to constrain thinking.
Ecourages main agent to run parallel searches.

[Copilot]

Pull request overview

Updates the project’s AI/LLM dependency set (notably pydantic-ai) and wires in per-agent “thinking”/reasoning controls via model_settings to better constrain when agents should (or shouldn’t) do extended reasoning.

Changes:

• Bump pydantic-ai and multiple related provider/SDK dependencies (OpenAI/Anthropic/Cohere/Mistral/etc.) via pyproject.toml, requirements.txt, and uv.lock.
• Add model_settings={'thinking': False} to several “tooling”/utility agents and set model_settings={'thinking': 'high'} for the main deck-construction agent.
• Adjust memory-writing behavior to raise ModelRetry (with logging) when too many related card UUIDs are provided, improving agent retry behavior.

Reviewed changes

Copilot reviewed 6 out of 8 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
uv.lock	Locks updated dependency graph reflecting the model/SDK bumps and new transitive deps.
requirements.txt	Exported requirements updated to match the new locked dependency set.
pyproject.toml	Raises minimum pydantic-ai version to >=1.87.0.
app/appai/services/agents/tools/memory_tools.py	Uses ModelRetry for invalid tool input and disables “thinking” for memory subagents via model_settings.
app/appai/services/agents/guardrails.py	Disables “thinking” for the guardrail agent to keep relevance checks lightweight.
app/appai/services/agents/filter_constructor.py	Disables “thinking” for filter construction agent to constrain reasoning/latency.
app/appai/services/agents/deck_theme.py	Disables “thinking” for daily theme generation agent.
app/appai/services/agents/deck_constructor.py	Sets high “thinking” for the main deck constructor while disabling it for classifier/replacement subagents; also updates the system prompt guidance.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.