embed_docsite: use git clone to download source files from repo

[josephjclark]

In embed_docsite, we download the docs source from github repos before processing and uploading to pinecone.

In github_utils right now, we download these files one at a time by downloading the index and then fetching each file.

I'd be willing to bet a shiny pound that git clone --depth 1 will be easier, faster, and more robust.

I've love to swap out the implementation here on a slow day. And run a quick test on the time-to-download before and after the change

[IZO-Ong]

I looked at this while working on embed_docsite's GitHub fetching and want to suggest an alternative to git clone --depth 1 where we use the Trees API (GET /repos/OpenFn/docs/git/trees/main?recursive=1) plus an on-disk cache keyed by ETag and per-file blob SHA.

Compared to git clone, the tree method avoids adding a git subprocess dependency, gets exact per-file change detection for free as {path: blob_sha} in the tree response instead of needing a git diff/sparse-checkout step to isolate docs/+adaptors/, and rate-limit failures come back as a typed ApolloError(429, ...) rather than a subprocess error.