Checking changes now takes too long because of sha retrieval

[kzu]

Previously, running an update was fairly quick, with the optimizations around using HEAD and ETag to conditionally and quickly determine if a full GET was needed.

Since the introduction of the SHA retrieval, however, things are much slower now, since we do a GH CLI call for (almost) all requests.

We should instead optimize and not retrieve the SHA if:

1. We have a matching ETag from request/response
2. We already have a SHA persisted