Add lint reporting and /fix-lint skill

[podmar]

Decisions worth noting

• run-lint.sh uses set -uo pipefail without -e so non-zero exit codes from Ruff and Pyright don't abort the script — lint failures are the expected case.
• The script runs both tools from backend/ via uv run rather than invoking binaries directly, so it honours the project's pinned versions without any PATH assumptions.
• The report is gitignored alongside pr-comments.md and pr-description.md — it's a local scratchpad, not a project artefact.
• /fix-lint distinguishes "clear errors" (auto-fixed) from "design decisions" (confirmed first) to avoid Claude silently adding # type: ignore or restructuring logic without review. The boundary is defined explicitly in the skill so the behaviour is predictable.
• Ruff and Pyright issues for the same file are surfaced together and cross-referenced, so a single root cause doesn't produce two separate fix passes.

What's next

Run run-lint against the current codebase to see what the baseline looks like.

[claude]

This PR adds developer tooling and looks good — no security, data isolation, or correctness concerns.

[claude]

Review Summary

Critical: None

Minor: None

This PR adds developer workflow tooling (lint runner + Claude Code command). No security, data isolation, or correctness issues. The changes are isolated to scripts and documentation — no backend code touched.

Cost: $0.12