feat: sessions files and sessions by-file subcommands

Feature Request

Add file-level tracking to the sessions subcommand group:

sessions files <session-id>

List files touched by a specific session, extracted from tool invocations in the JSONL data.

Extraction targets:

• Read tool -> tool_input.file_path
• Edit tool -> tool_input.file_path
• Write tool -> tool_input.file_path
• MultiEdit tool -> tool_input.file_path
• NotebookEdit tool -> tool_input.notebook_path
• Glob tool -> tool_input.path
• Grep tool -> tool_input.path

Output:

• Categorize as "read" vs "written" (Edit/Write/MultiEdit/NotebookEdit = written)
• Support --json for programmatic use
• Default: human-readable table

sessions by-file <file-path>

List sessions that touched a given file path (substring match).

Output:

• Session ID, topic, read/write mode
• Support --json

Context

terraphim-agent already parses JSONL session files for sessions search, sessions list, and sessions stats. This extends extraction to the tool_input fields for file path tracking.

A temporary bridge script exists at scripts/session-files.py in the cto-executive-system project. This upstream implementation would supersede it.

Proposed CLI

terraphim-agent sessions files <session-id> [--json]
terraphim-agent sessions by-file <file-path> [--json]

Start by reading the existing sessions search, list, and stats implementations that parse JSONL session files, then compare them with scripts/session-files.py in the cto-executive-system project. Trace how tool_input fields are extracted and how CLI output supports human-readable and JSON formats. Done means both proposed subcommands report the requested read/write file and session information with --json support.