Improve list-repos compatibility and performance

[marcleblanc2]

Summary

• query Sourcegraph version, user permissions, and GraphQL schema once at startup, then mark unsupported or admin-only CSV fields explicitly
• reuse HTTP connections, bound concurrent work, validate incomplete responses, and use shared retry backpressure
• retain complete output while adding sorted per-run CSVs, optional skipped-file trigram metrics, and 90%-of-request search timeouts
• improve retry resilience, error summaries, empty-repository commit counts, and end-of-run issue totals
• document generated CSV output and remaining optimization work

Verification

• uv run ruff format --check list-repos.py
• uv run ruff check list-repos.py
• uv run pyright list-repos.py
• uv run python -m py_compile list-repos.py
• npx --no-install markdownlint-cli2 AGENTS.md CSV_SCHEMA.md dev/TODO.md
• focused checks for error truncation, GraphQL message limits, empty repositories, and issue totals
• live one-repository run against sourcegraph.sourcegraph.com with stats, skipped-file details and metrics, commit counting, and search
• full benchmark: 1514.404s versus 2224.845s baseline (31.9% faster)

[github-actions]

Open the Diff Tour for this PR in Sourcegraph