fix(GCS+gRPC): correctly resume ranged reads

[coryan]

Fixes #14494

---

This change is 

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 93.59%. Comparing base (eb2cacc) to head (d6b4644).

Additional details and impacted files

@@            Coverage Diff             @@
##             main   #14495      +/-   ##
==========================================
- Coverage   93.60%   93.59%   -0.01%     
==========================================
  Files        2316     2316              
  Lines      207319   207348      +29     
==========================================
+ Hits       194054   194077      +23     
- Misses      13265    13271       +6     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[dbolduc]

Q: why should we even allow ReadFromOffset and ReadRange together? It is only well defined if offset == range.begin, no?

[coryan]

Q: why should we even allow ReadFromOffset and ReadRange together? It is only well defined if offset == range.begin, no?

You could argue that the intersection of ReadFromOffset and ReadRange is reasonably well defined if offset < range.end.

Cleverly (see the air quotes?) the resume loop only sets ReadFromOffset, when/if the application also sets ReadRange we merge both. Not the cleanest design, but a bit tricky to (a) prohibit, given the APIs we do expose, or (b) fix without risking a break for JSON.

Let me consider your proposed fix separately.