feat: Add benchmark CLI command for MCP tool timing (closes #63)

[evansenter]

Summary

• Adds session-analytics-cli benchmark command to measure response times for all 26 MCP tools
• Calculates timing statistics (median, p95, p99) and flags slow queries (>5s)
• Updates CLAUDE.md to remind developers to add new tools to benchmark infrastructure

Test plan

• session-analytics-cli benchmark runs successfully against production DB
• All 303 tests pass including new benchmark tests
• make check passes (format, lint, tests)

Output example

Benchmark Results (3 iterations per tool)
Total tools: 26
Slow tools (>5s): 0

TOOL                                MEDIAN        P95        P99     STATUS
-----------------------------------------------------------------------------
get_insights                        1.234s     1.456s     1.456s        OK
analyze_trends                      0.145s     0.234s     0.234s        OK
get_tool_frequency                  0.045s     0.048s     0.048s        OK

🤖 Generated with Claude Code

[claude]

Prompt: evansenter/dotfiles/.../claude-review.md

Code Review

Summary

This PR adds a session-analytics-cli benchmark command that measures response times for 26 MCP tools, calculating median/p95/p99 statistics and flagging slow queries (>5s). It includes proper documentation updates and unit tests.

Issues Found

Critical

None

Important

None

Suggestions

• cli.py:564 - The error formatting line has inconsistent column widths. When there's an error, the status column shows the error message instead of using a dedicated error display. Consider using {err_msg:>10} for consistency with the status column, or showing "ERROR" in the status column and the full error message on a separate line.

• cli.py:1036-1038 - The p95/p99 percentile calculation with 3 iterations will always return the same value (the max). With n=3, int(3 * 0.95) = 2 and int(3 * 0.99) = 2, so both p95 and p99 will be times[2] (the max). Consider documenting this limitation or recommending a higher --iterations value (e.g., 10-20) for meaningful percentile statistics.

Verdict

REQUEST_CHANGES - Found suggestions for improving error display formatting and documenting percentile calculation limitations with low iteration counts.

---

Automated review by Claude Code

[evansenter]

Feedback Addressed

Implemented

• [Suggestion] Added help text note for --iterations: "use 10+ for meaningful p95/p99"

Skipped

• [Suggestion] Error display formatting - Current inline display is readable and shows error directly where users would look

[claude]

Prompt: evansenter/dotfiles/.../claude-review.md

Code Review

Summary

This PR adds a session-analytics-cli benchmark command that measures response times for 26 MCP tools with configurable iterations, calculating median/p95/p99 statistics and flagging slow queries (>5s). The implementation includes comprehensive tests and documentation updates.

Previous Feedback Status

Both suggestions from the previous review have been addressed:

• Error display formatting - Skipped with valid reasoning (inline display is readable)
• p95/p99 iteration note - Implemented (help text now says "use 10+ for meaningful p95/p99")

Issues Found

Critical

None

Important

None

Suggestions

None

Verdict

APPROVE - Code is well-structured, follows project conventions, includes tests, and addresses all requirements from issue #63. Previous feedback has been appropriately addressed.

---

Automated review by Claude Code