doctor: report every agent, not just Codex

[zzet]

The runtime half of doctor described one agent. The invocation log has been partitioned per agent since it learned to record one, and the finding rules take the agent as a parameter — only the collection and the rendering singled Codex out. So a machine running Claude Code got a detailed report about the agent it uses less and nothing about the one it uses most.

Doctor now walks a list of agents, each contributing install state, its own slice of the hook log, adoption from its own transcripts, and its own findings, tagged so a reader can tell whose is whose. Adding an agent is now the cost of one probe entry; the rules, the log partition, and the renderer are already generic.

Claude Code gains the two readers Codex already had. Inspect covers the hooks in settings.local.json, the MCP stanza in ~/.claude.json, and the rule block in ~/.claude/CLAUDE.md, reusing this package's own path helpers so a file the adapter writes and a file the reporter looks for cannot drift. The transcript scanner reads assistant messages' tool_use parts; a native Read/Grep/Glob counts in the same denominator as a shell read, because it is the same competing behaviour.

Only Codex carries the trust gate, so only Codex gets the /hooks remedy — the rules already handled that distinction, it just had no second agent to prove it against.

Why this is a separate PR

This commit was pushed to fix/doctor-mixed-attribution-and-redact about thirty minutes after #353 had already merged. A merged PR does not reopen on a push to its head branch, so the change was never proposed anywhere and never reached main. Cherry-picked here onto current main, unchanged — identical patch-id to the original.

Verification

go vet and go test pass for both touched packages:

ok  github.com/zzet/gortex/internal/agents/claudecode  0.904s
ok  github.com/zzet/gortex/internal/doctor             0.297s