fix(mcp): dedicated timeout for initialize handshake

[Svagtlys]

Closes #101

Summary

The MCP initialize handshake previously borrowed the steady-state per-request timeout (request_timeout_seconds) as its start-up budget, so callers who tightened that knob strangled slow-to-boot servers (and flaked tests). This PR gives the handshake its own budget:

• New setting: McpSettings.initialize_timeout_seconds (default 30.0, env OCTAVE_MCP_INITIALIZE_TIMEOUT_SECONDS) in octave/mcp/config.py.
• Client swap: _await_initialize()'s fail_after and connect()'s timeout error message now read the dedicated knob; _run() and the steady-state budget are untouched. Zero-impact for default settings (both default to 30 s).
• Docs: OCTAVE_MCP_INITIALIZE_TIMEOUT_SECONDS row added to the MCP env-var table in docs/DEVELOPMENT.md.

Commits

• feat(mcp): add initialize_timeout_seconds setting
• fix(mcp): run initialize handshake under its own timeout budget
• docs(mcp): document OCTAVE_MCP_INITIALIZE_TIMEOUT_SECONDS
• docs(specs): design and plan for dedicated initialize handshake timeout

Test evidence

4 new tests (TDD: verified failing before implementation, passing after):

• test_settings_default_initialize_timeout / test_settings_env_override_initialize_timeout — config default + env override
• test_initialize_timeout_uses_dedicated_budget — silent peer; error cites 0.05 (dedicated budget), not 1.0 (request budget)
• test_slow_initialize_survives_tight_request_timeout — regression: peer boots in 0.2 s under request_timeout=0.05; connect still succeeds on the 30 s initialize budget

Gates:

• uv run pytest — 281 passed
• uv run ruff check . — All checks passed
• uv run mypy src — Success: no issues found in 48 source files

Checklist

• Implementation complete
• Tests passing
• Documentation updated