The two GET steps (get-ref, get-commit) used resp.statusText, which is
often empty on HTTP/2. Now they read resp.text() like the other 4 throw
paths so every error message includes GitHub's response body for
debugging.
Plus a test assertion for calls[2] in the Git Data API path so a
transposition of GET ref / GET commit would be caught.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Blobs ≤ BLOB_THRESHOLD_BYTES (900 KB) take the Contents API path
(same as writeFile). Larger blobs use the Git Data API: POST blob,
GET ref + commit, POST tree (with base_tree), POST commit, PATCH ref.
Tests cover both paths plus error propagation.
getBlob/deleteBlob are thin wrappers over readFile/deleteFile.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
