feat: LLM-powered test analysis pipeline (v0.2.0)

Add Claude-powered failure analysis, root cause detection, and report
generation to the LISA MCP server.

New modules:
- lisa_mcp/tools/log_collector.py  — memory-safe LISA run log scanner
  (seek-based tail reading, 256 KB/file cap, error context extraction)
- lisa_mcp/tools/llm_analyzer.py   — Anthropic API calls with tool_use
  for structured JSON output (FailureAnalysis + RunAnalysisSummary)
- lisa_mcp/tools/report_generator.py — self-contained HTML + Markdown
  report rendering (no CDN, works offline)

New MCP tools (13 → 17 total):
- analyze_test_run_with_llm    — analyze all failures in a run
- analyze_failure_root_cause   — deep-dive on a single failure
- generate_analysis_report     — full HTML+MD report from results
- run_and_analyze              — end-to-end: run tests → analyze → report

New models in models.py:
- RootCauseCategory (12 categories: kernel_panic, network_timeout, ...)
- FailureSeverity   (critical / high / medium / low)
- FailureAnalysis   — per-test: category, description, fix, log lines
- RunAnalysisSummary — run-level: health, patterns, priorities, summary
- AnalysisReport    — complete report data model

Documentation:
- docs/llm-analysis.md — full guide including pipeline diagram,
  per-step examples, cost guidance, and CI/CD integration patterns

Dependencies:
- anthropic>=0.40.0 added to pyproject.toml

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

Author: kkkashan