Add pytest-perf-analysis agent skill

[ehelms]

Why are you introducing these changes? (Problem description, related links)

The .agents/skills/ directory exists but has been empty. This adds a first
skill for analyzing pytest performance — a repeatable workflow that was
previously done ad-hoc: paste timing output, read the associated fixture code,
identify anti-patterns, and produce ranked recommendations.

What are the changes introduced in this pull request?

• Add .agents/skills/pytest-perf-analysis/SKILL.md — skill definition with
  trigger description, input handling (pasted output, file path, or live run),
  analysis process (extract slow items → find fixture definitions → apply
  anti-pattern checklist → check fixture dependency graph → identify
  inherently-slow tests), and a fixed output format with root causes and
  recommendations grouped by impact
• Add .agents/skills/pytest-perf-analysis/references/fixture-antipatterns.md
  — reference file documenting all four pytest timing output format variants
  (--fixture-durations table, --durations, combined report, JUnit XML) and
  five named anti-patterns with before/after examples and expected speedups:
  function-scoped expensive remote resources, overly broad task/async waits,
  O(n×m) teardown loops, multiple runtime startups per test, and inherently
  slow tests mixed with fast tests

How to test this pull request

Steps to reproduce:

• Run the foremanctl test suite with timing flags and paste the output to
  Claude with this skill active — it should produce a structured root-cause
  analysis and ranked recommendations without manual prompting

Checklist

• Tests added/updated (if applicable)
• Documentation updated (if applicable)

[ehelms]

I generated this as part of trying to analyze slow tests and identify a way to do this systematically in the future. I used this in conjunction with #666 to do both. I don't see this as a "continual" skill, but one that can be handy from time to time.