feat(mcp): add FastMCP server for LLM agent integration

[federicogmz]

Description

Implements native Model Context Protocol (MCP) server support (shalstab-mcp) in shalstab-python. This exposes deterministic, physically-based slope stability tools directly to frontier AI agents (Claude Desktop, Claude Code, OpenAI Agents SDK).

Features

• FastMCP Analytical Server (shalstab/mcp_server.py):
  • compute_infinite_slope_stability: Analytical Coulomb / Skempton-Delory infinite slope equilibrium with SHALSTAB 4-class stability categorization across dry, saturated, and steady-state regimes.
  • diagnose_topographic_wetness: Montgomery & Dietrich (1994) steady-state wetness index, critical recharge ratio, $\log_{10}(q/T)$ and natural $\ln(q/T)$ failure thresholds.
  • Pure Python analytical core without GIS dependencies at module startup (sub-second launch, 100% clean stdio).
• Lazy Loading: __getattr__ in shalstab/__init__.py delays heavy imports until Analyzer is accessed.
• Dual MCP Protocol Support: Compatible with both mcp 1.x (FastMCP) and mcp 2.x (MCPServer).
• Comprehensive Unit Tests (tests/test_mcp.py): 13 unit tests covering analytical Factor of Safety, unconditional stability bounds, and MCP server factory / tool execution.
• Documentation & Quickstarts: Full guides for Claude Desktop and OpenAI Agents SDK in README.md.

Verification

• 51/51 unit tests passed across repository (pytest tests/ -v).
• Parity verified with ruff check . and ruff format --check . on Ruff 0.16.10.
• Reviewed and approved by Advisor (Claude Opus 5.5).