Remove Cursor from router client documentation

[cursor]

Symptom

Public router docs and the local seed helper still advertised Cursor setup (OpenAI base URL override), despite no installer support and a poor client experience.

Change

• Remove Cursor sections from README, install docs, and npm package README.
• Drop Cursor setup output from cmd/seed.
• Remove Cursor column from the directive support table in install docs.

Validation

• Interactive installer target picker already lists Claude Code, Codex, opencode, and pi only.
• go vet ./cmd/seed/... passes.

Fixes W-1683

  

[devin-ai-integration]

Thanks for this — docs-only cleanup, and it lines up with how the installer actually behaves, so nothing to rewrite here. I checked the premise rather than taking it on faith: install/install.sh really does only carry claude / codex / opencode / pi targets (target="claude" default plus the --codex/--opencode/--pi flags), so the Cursor sections were advertising a path the installer never owned. The cmd/seed trim is the same story, and go vet ./cmd/seed/... plus the installer suite stay green.

Conventions-wise it's in line with the repo: docs live where root AGENTS.md puts them, no production code paths touched, and the directive table edit keeps the same column-per-client shape the rest of install/README.md uses.

Two leftovers you may want to fold in — neither is a rules violation, just references the diff makes stale:

• docker-compose.yml (the seed service comment) still says the one-shot seeder "prints the raw token with paste-ready configuration for Claude Code and Cursor", which is no longer true after the cmd/seed/main.go change.
• install/directives.tsv still declares a cursor column with every row set to manual, install/registry.sh still destructures it, and install/tests/registry_test.sh asserts cursor = manual with a comment explaining that the registry must say so "rather than implying this installer can write Cursor's settings." The paragraph in install/README.md that documented that column is exactly what this PR removes, so the machine-readable registry now models a client the docs no longer mention. Either is a defensible end state — keep the column and keep a one-line doc note, or drop the column too — but right now they disagree.

Note the one red check (review) is the known comment_length_review.yml limitation with bot-initiated runs, not anything in this diff.

The one caveat worth a human eye, since it's a product call and not a code one: internal/proxy/client_identity.go still detects and labels Cursor traffic (ClientAppCursor), so Cursor users pointed at the router are still first-class in telemetry — this PR only stops us advertising the setup. Worth confirming that's the intent before it leaves draft.