fix(server): show Codex and Claude thinking traces in V2 adapters

[SunkenInTime]

What Changed

Ports #11784 (feat(chat): show provider thinking traces) into the V2 orchestration adapters, which the v2 rewrite left behind when it deleted the v1 ProviderRuntimeIngestion / ClaudeAdapter / CodexAdapter files.

• CodexAdapterV2: handles item/reasoning/summaryTextDelta and item/reasoning/textDelta via a second makeCodexAgentMessageDeltaCoalescer, emitting a kind: "reasoning" node + type: "reasoning" turn item (streaming: true while deltas arrive). Indexed parts (summaryIndex / contentIndex) are joined with \n\n, mirroring main's ingestion. item/completed for a reasoning item finalizes from item.summary / item.content; flushTurn runs at turn end. Summary text keeps the native item id; raw chain-of-thought (when Codex emits it) is a sibling item with id ${itemId}:raw.
• ClaudeAdapterV2: thinking content blocks on a completed assistant SDK message become one completed reasoning item (${uuid}:thinking), emitted before the assistant text so it orders ahead of the answer. Query options now set showThinkingSummaries: true unless the thinking option is off or --thinking-display omitted is passed (same rule as main's shouldRequestClaudeThinkingSummaries).
• apps/web/session-logic.ts: workEntryIndicatesToolNeutralStatus no longer treats a reasoning entry that has text as neutral. Without this, the completed reasoning items are persisted but filtered out of the expanded Worked group (workEntryIsVisibleInGroup), so the trace was invisible even after the adapter fix. This also un-hides completed Cursor/OpenCode/Pi/ACP V2 reasoning items, which were already emitted but hidden.

Focused tests: Codex replay transcript (summary + raw deltas, indexed separation, completion, ordering vs. final answer), Claude thinking-block projection + showThinkingSummaries option rule, and the web neutral-status rule.

Why

On t3code/codex-turn-mapping, Codex and Claude thinking traces are dropped: CodexAdapterV2 registered no item/reasoning/* handlers (despite advertising streamsReasoning: true) and ClaudeAdapterV2 ignored thinking blocks. Main's fix (052c7ae) is an ancestor of this branch but only touched v1 files that the V2 rewrite removed, so it merged as a no-op. The V2 contract (reasoning node/turn item) and the UI mapping (reasoning -> "Thinking" work entry) already existed; only the adapters (and the group visibility rule) needed the port.

Reproduction

No Codex/Claude credentials were available, so the repro drives the real web app against the repo's mock Codex app-server (apps/server/src/provider/testFixtures/codexCollabMockPeer.sh), which replays scripted notifications on turn/start.

1. Isolated state per worktree (vp run dev --home-dir <worktree>/.t3, never ~/.t3/userdata); providers.codex.binaryPath in that .t3/userdata/settings.json pointed at codexCollabMockPeer.sh; server started with T3_CODEX_COLLAB_SCRIPT=<script.json>.
2. Script notifications: item/started (reasoning), two item/reasoning/summaryTextDelta (summaryIndex 0 and 1), item/completed (reasoning with both summaries), then an agentMessage final answer.
3. Pair the browser, add a scratch project, start a Codex thread, send a prompt, wait for the turn to complete, expand "Worked".

• Expected: a Thinking row with "Weighing the options" / "Choosing the shortest path" above the final answer.
• Actual (base f08294a11 (pre-rewrite tip of t3code/codex-turn-mapping)): only the final answer; no Thinking row; the V2 DB contains zero reasoning items.

UI Changes

Before (base f08294a11 (pre-rewrite tip of t3code/codex-turn-mapping))	After (this branch)

V2 item inspector after the fix, showing reasoning-1 completed with both summaries:

Verification

• vp test run on CodexAdapterV2.test.ts, ClaudeAdapterV2.test.ts (183 passed) and apps/web/src/session-logic.test.ts.
• tsc --noEmit for apps/server and apps/web (only pre-existing @cursor/sdk type errors in Cursor files before pnpm install; clean after).
• vp check on the changed files (only pre-existing unused-variable warnings).
• Computer-use before/after in the web app as described above; reasoning survives reload and appears in the V2 inspector.

Verification blockers: not tested against live Codex or Claude CLIs (no credentials on the test machine), so real streaming timing, raw textDelta content from Codex, and Claude thinking blocks from the real SDK were only covered by unit tests. Desktop/mobile not exercised; mobile renders V2 work entries through packages/client-runtime and was not changed.

Checklist

• This PR is small and focused
• I explained what changed and why
• I included before/after screenshots for any UI changes
• I included a video for animation/interaction changes (n/a)

Written by Devin (Claude) via the Devin harness.