bug(conversations): resuming in place from the picker fails with no message #1241

Found by the read-only C3 hunt (quality loop, 2026-09-25, temp/quality-loop/hunt-c3.md). The failure sequence is from the code; the loop has not yet verified it end to end. Verify before fixing.

ConversationsPicker.tsx:103-113 closes the picker, then awaits workspace.replaceSession(...). The callers use void resume(row), with no catch anywhere in the chain.

• A spawn failure (e.g. a CLI removed since the transcript was written) rethrows sessionSpawnErrorMessage. It becomes an unhandled rejection, which is only journaled.
• replaceSession can also return undefined with no message: no command target, missing meta, or canCommit failing.

The user sees nothing change. By contrast, the new-tab branch toasts on failure (tab.ts:54-55), and builtInMcpReload.ts:74-79 wraps the same call with a pane toast.

Expected: a pane toast with the reason, as builtInMcpReload does.