make it more likely that a STREAM frame is bundled with the FIN

[marten-seemann]

Fixes #2435.

Currently, on Stream.Write(), we save the slice in dataForWriting, and pop STREAM frames from that slice. Write only returns after all the data has been popped in STREAM frames.

This PR changes this logic: As soon as dataForWriting becomes smaller than a certain size x, it's copied to a STREAM frame, which can the be popped the next time we assemble a packet. This allows us to return Write() when all data but x bytes have been sent out. When the user now calls Close(), this is much more likely to happen before we popped that last STREAM frame, allowing us to set the FIN bit on that frame (and avoiding sending out an empty STREAM frame with FIN set in a separate packet).

[codecov-io]

Codecov Report

Merging #2504 into master will decrease coverage by 0.02%.
The diff coverage is 97.67%.

@@            Coverage Diff             @@
##           master    #2504      +/-   ##
==========================================
- Coverage   86.10%   86.08%   -0.02%     
==========================================
  Files         122      122              
  Lines        9490     9539      +49     
==========================================
+ Hits         8171     8211      +40     
- Misses        982      988       +6     
- Partials      337      340       +3     

Impacted Files	Coverage Δ
send_stream.go	92.96% <97.67%> (+0.01%)	⬆️
internal/handshake/header_protector.go	75.38% <0.00%> (-7.67%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 81daa8a...9905774. Read the comment docs.