chore(claude): retire claude platform plugin

[muzamilkm]

Summary

Retires our claude (Claude Platform) plugin. This is phase 1b of the Claude convergence plan (claude-code-upstream-convergence-plan.md) and builds on #79, which brings in upstream's claude_code. Edly's Claude usage is covered by OTel, so the Platform setup is deleted rather than migrated. Nothing else changes: claude_otel, cursor, codex, the shared ai domain package and every dashboard are untouched.

Plugin removed

• Deletes backend/plugins/claude/ (28 files) and config-ui/src/plugins/register/claude/.
• Drops its ClaudeConfig registration and its table_info_test.go import and FeedIn.
• connections.tsx: CLAUDE_PLUGIN now points at claude_code, so the "Claude Code OTel" card sits next to the Claude Code tile. Without this the card falls through to the end of the A–N list.

Core migration 20261002000001 (retire claude platform plugin)

Clears everything the plugin can leave behind, by exact name only. Every step is a no-op when there is nothing to remove, so it runs the same whether the connection was deleted first or not:

• Drops _tool_claude_connections (which holds the encrypted Admin key), _tool_claude_scopes, _tool_claude_scope_configs, _tool_claude_usage and _raw_claude_usage. Drops are IF EXISTS; _raw_claude_usage only exists once a collection has run.
• Deletes the plugin's ai_activities rows by _raw_data_table = '_raw_claude_usage'. Matching on provider = 'claude' would be wrong: claude_otel writes that provider too.
• Deletes its _devlake_collector_latest_state and _devlake_subtask_states rows.
• Deletes claude entries from _devlake_blueprint_connections and _devlake_blueprint_scopes. A blueprint that still referenced the plugin would otherwise fail every run with an unknown-plugin error.
• Deletes the plugin's 3 _devlake_migration_history rows, matched on version and name, because cursor and codex reuse 2 of those version numbers. Reverting this PR then reruns the plugin's own migrations and recreates its tables.

There is no _tool_claude_% pattern anywhere: _tool_claude_code_otel_* (live data) and upstream's _tool_claude_code_* share that prefix.

Test plan

• Replicated a deploy with leftovers on the local stack. Seeded claude data everywhere the migration cleans: a connection, scope, scope config, 2 usage rows, a _raw_claude_usage table with a row, 2 ai_activities rows, collector and subtask state, and a blueprint connection and scope. Then rebuilt devlake and config-ui with FORCE_MIGRATION=true:
  • exactly 1 script ran (retire claude platform plugin), with no errors, and the server came up;
  • the 5 tables are gone and every seeded row is deleted;
  • a row-count diff of all 517 tables against the pre-seed snapshot shows only the dropped tables, migration history 447 → 445 (3 claude rows out, 1 in) and the usual restart lock row;
  • the 85 OTel provider = 'claude' activities, all _tool_claude_code_otel_* tables and the cursor/codex history rows that share version numbers are unchanged.
• The built image ships claude_code and claude_otel and no claude plugin.
• go vet on the new migration, and go test for claude_code, claude_otel and domainlayer/ai, pass. The full table_info_test needs libgit2 1.3.0 and runs in CI.
• Config UI: tsc, eslint and prettier pass, and no imports of register/claude remain.

Notes for reviewers

• Do not merge until Edly-Claude is gone from prod: delete the Edly-Claude project (which removes blueprint 10) and connection claude#1, and have that Admin key revoked in the Anthropic Console. The migration cleans up any leftovers either way, but the key should be revoked regardless.
• Rollback: revert the PR. The plugin's migrations rerun and recreate empty tables; there is no data to restore.
• Merge after feat(claude-code): cherry-pick upstream claude_code plugin #79. This branch is stacked on it, so the diff shows both until feat(claude-code): cherry-pick upstream claude_code plugin #79 merges.