Fix mutable default previous_run parameters across assistants

[DavidJBianco]

Motivation

• The code used shared mutable defaults like previous_run: list = list() in multiple assistant entrypoints, which can retain state across invocations and leak prior session context in multi-user or server scenarios.
• This is a security-sensitive information disclosure risk when agents are invoked by the Streamlit/API or MCP server paths.

Description

• Replaced mutable defaults previous_run: list = list() with safe optional defaults previous_run: Optional[list] = None in assistant entrypoints.
• Added Optional to type imports where required so annotations remain explicit and valid.
• Changes were applied to peak_assistant/able_assistant/__init__.py, peak_assistant/data_assistant/__init__.py, peak_assistant/hypothesis_assistant/hypothesis_refiner_cli.py, peak_assistant/planning_assistant/__init__.py, and peak_assistant/research_assistant/__init__.py to remove the shared-list pattern while preserving existing function signatures and behavior.

Testing

• Ran python -m compileall -q peak_assistant to verify the package compiles, and it succeeded.
• Verified there are no remaining occurrences of the pattern previous_run: list = list() under peak_assistant using a repository search, and the check returned no matches.

---

Codex Task

[DavidJBianco]

Merged into dev via #88 (batch-d-refiner-fixes, commit 50bdee2).