Fwrite gzip

[philippechataignon]

Closes #2016
Closes #2974
Closes #2612

I reopen a PR to discuss about its contents. Its a complete rewrite of #3278 : it doesn't use gzopen/gzwrite/gzclose because only one thread compress and that's slow.

In fwrite, output file is always open in binary mode. The idea is to take data from uncompressed buffer in normal case and to compress it and take data from compressed buffer when gzip is active. The code for managing file output is quite the same in normal and gzip mode.

The main difference concern the buffers :

• there are no realloc in this version : if the buffer is too small, fwrite stops and suggests to increase the buffMB parameter. Its too difficult to realloc the compressed and the uncompressed buffer. But, in fact, actual buffer policy doesn't work very good despite realloc. See fwrite crashes for tables with large columns #2974 [request] print informative error if fwrite uses more memory than allowed #2612.
• at the beginning, fwrite compute the max line length. Actually it tries to write in the buffer and add the ending position. But if the buffer is too small, its too late : buffer overflow occurs. Thats what happens in fwrite crashes for tables with large columns #2974. In this PR, I use strlen for string/factor and an maximum length integer by type (9 for 32 bits, 19 for int 64 for example). It's not a problem to overestimate this line.

Thanks for any advice on this PR.

[codecov]

Codecov Report

Merging #3288 into master will increase coverage by 0.29%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #3288      +/-   ##
==========================================
+ Coverage   96.71%   97.01%   +0.29%     
==========================================
  Files          66       66              
  Lines       12424    12475      +51     
==========================================
+ Hits        12016    12102      +86     
+ Misses        408      373      -35

Impacted Files	Coverage Δ
src/fwrite.c	97.59% <100%> (+6.58%)	⬆️
R/fwrite.R	98.24% <100%> (+1.94%)	⬆️
src/fwriteR.c	98.02% <100%> (+3.11%)	⬆️
src/nafill.c	100% <0%> (ø)	⬆️
src/freadR.c	96.28% <0%> (+0.12%)	⬆️
src/uniqlist.c	98.98% <0%> (+0.5%)	⬆️

---

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 c30e161...7665a8d. Read the comment docs.

[mattdowle]

Fantastic PR -- thanks @philippechataignon.