Fix issues with voyager

[RustedBones]

No description provided.

[codecov]

Codecov Report

Merging #5034 (e3e637a) into main (fb3c27f) will decrease coverage by 0.01%.
Report is 6 commits behind head on main.
The diff coverage is 78.57%.

❗ Current head e3e637a differs from pull request most recent head 388f1fa. Consider uploading reports for the commit 388f1fa to get more accurate results

@@            Coverage Diff             @@
##             main    #5034      +/-   ##
==========================================
- Coverage   63.12%   63.12%   -0.01%     
==========================================
  Files         286      286              
  Lines       10736    10743       +7     
  Branches      764      791      +27     
==========================================
+ Hits         6777     6781       +4     
- Misses       3959     3962       +3     

Files	Coverage Δ
...scala/com/spotify/scio/extra/voyager/Voyager.scala	66.66% <100.00%> (ø)
.../scio/extra/voyager/syntax/SCollectionSyntax.scala	57.14% <100.00%> (+7.14%)	⬆️
...potify/scio/bigquery/types/ConverterProvider.scala	100.00% <ø> (ø)
...m/spotify/scio/bigquery/types/SchemaProvider.scala	95.91% <100.00%> (-0.17%)	⬇️
...com/spotify/scio/bigquery/types/TypeProvider.scala	47.94% <ø> (ø)
...la/com/spotify/scio/bigquery/types/MacroUtil.scala	90.47% <0.00%> (-0.44%)	⬇️
.../src/main/scala/com/spotify/scio/ScioContext.scala	82.52% <66.66%> (-0.40%)	⬇️

... and 1 file with indirect coverage changes

[clairemcginty]

out of curiosity, how does Dataflow render this job graph? I guess it adds another root-level transform for .parallelize() that feeds into the Voyager SCollection?

[RustedBones]

yes:

I'm just getting rid on the extra map from reifyAsIterableInGlobalWindow that can trigger the iterableCoder forcing the side input collection to fit in memory. Here we directly access the side input that should be able to stream records from the view