Make entire explain work with partial-clone

[Soph]

Problem

entire explain <checkpoint-id> failed with "no checkpoint or commit found" against a checkpoint_remote-backed setup with filtered_fetches: true. The checkpoint was on the remote and discoverable in the local tree, but the metadata blob was absent locally — and go-git's Tree.File() returns ErrFileNotFound for missing blobs, which the v1 / v2 read paths treated as "checkpoint doesn't exist." Even when it eventually returned correct data, it could take 10+ seconds on a cold cache with zero progress feedback — easy to mistake for a hang.

Fix — correctness

Three coordinated changes:

1. Configure blob fetcher on both stores. v1Store had SetBlobFetcher but explain wasn't calling it. Added the same to V2GitStore and wired both.
2. Wrap v2 reads in FetchingTree. v1's read path already used FetchingTree (with cat-file fallback for partial-clone-filtered blobs). Mirrored that for v2's ReadCommitted / ReadSession* methods.
3. Use git fetch-pack for blob SHAs. Porcelain git fetch <url> <hash> enforces partial-clone integrity checks that reject blob-only responses; plumbing skips those checks. Switched FetchBlobs accordingly.

Plus: FetchingTree.File now tries git cat-file BEFORE the network fetch — saves multi-second round-trips when a blob is on disk but invisible to go-git's storer (typical partial-clone state).

Fix — progress feedback

Wrapped the entire pre-output pipeline with two sequential spinners on stderr:

• ⣾ Loading checkpoints during newExplainCheckpointLookup (the slow ListCommitted walk that reads every checkpoint's metadata.json via git cat-file)
• ⣾ Loading checkpoint <id> during prefetchCheckpointBlobs (including its cat-file -e analysis loop), the actual fetch, summary + session reads, and getAssociatedCommits git log walk

Same Braille frames as the activity TUI for visual consistency. 250ms initial delay so warm/fast runs stay silent — no flicker. Stops strictly before any write to stdout so spinner frames and output never interleave. Suppressed entirely when stderr isn't a terminal (CI, redirected output, agent subprocesses).

Tests

Two new integration tests reproduce the partial-clone state in a fresh TempDir clone (file:// URL + uploadpack.allowFilter=true on the bare so --filter=blob:none is actually honored — the local-path transport ignores filters) and verify explain succeeds:

• TestExplain_CheckpointSucceedsAfterTreelessFetch (v1)
• TestExplain_CheckpointV2SucceedsAfterTreelessFetch (v2)

Both assert at least one metadata blob is genuinely missing locally before running explain — without that pre-condition the tests could silently pass against a fully-cloned repo. Verified for regression: stashing the v2 fix makes the v2 test fail with the exact original symptom (failed to read checkpoint: checkpoint not found).

Performance

• Cold run (blobs missing locally): ~1-2s — one batched fetch-pack call covers all blobs in the checkpoint subtree, no per-blob round-trips
• Warm run (blobs on disk): sub-second — cat-file reads local blobs directly, no network

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit ae4b4e5. Configure here.}