fix(runtimed): loop for automerge sync ack in sync_to_daemon

[rgbkrk]

sync_to_daemon previously read exactly one frame after sending the sync message. If a Broadcast frame arrived before the AutomergeSync ack, the broadcast was silently dropped and the ack was left unprocessed in the stream buffer — leaving peer_state stale.

Now loops until the ack arrives (or the 500ms deadline expires), queuing any Broadcast frames in pending_broadcasts for later delivery.

Also adds diagnostic logging to ExecuteCell when a cell is not found, including the doc's cell count and IDs.

Addresses these flaky CI failures:

FAILED test_cell_created_by_one_visible_to_other - Timed out waiting for cell with source visible to s2
FAILED test_interleaved_stdout_stderr_separate - Cell not found in document
FAILED test_uv_inline_deps_trusted - uv:inline requires a prepared environment (was it created?)

All three share the same root cause: sync mutations not propagating because a Broadcast frame ate the AutomergeSync ack.

Supersedes #693 (same fix, rebased onto current main after fractional indexing landed).

PR submitted by @rgbkrk's agent Quill, via Zed