Use BGZF* in cram_index_build() instead of zfp*

[jmarshall]

Also fixes unchecked-write-error bugs in index writing.

The routines in zfio.c are now unused so can be removed. As a happy byproduct, HTSlib no longer uses fseeko(), so no heroics are needed for platforms where that function is not available.

[jkbonfield]

Thanks John.

[jmarshall]

If I'd realised there would be as little to change as that, I'd have got around to it years ago 😄

[jkbonfield]

Quite. "Getting around to it" is always a tricky thing. I decided to keep going with it noticing vlen.h next to the removed zfio.h so culled the reason for that existing too. :-)

The io_lib legacy lives on still in places, but it's all an incremental improvement.

[jmarshall]

Yep… nice.

My plan for that corner was to replace cram_io.c's open_path_mfile() and find_path() calls with HTSlib's wider facilities that by now have (almost?) expanded to do what's actually needed for reference path handling. At that point, mFILE.[ch], open_trace_file.[ch], and vlen.[ch] all evaporate.

[jkbonfield]

There is still #417 which is on my radar to check out. This may well provide most of the functionality needed as it's been written with the same need in mind - faking up file IO from a memory buffer.

Although to be frank the reference handling code was rather "organic" during development and is a rats nest in need of a complete rewrite. It would be useful outside of CRAM too, for things like mpileup and tview. That's a bit of a longer project needing a rainy month.

[jmarshall]

hFILE fixed buffers, introduced for data: URLs but over the top just for those, and var-args hopen were designed with that memory buffer PR in mind. I've written up my past plans for such functionality over on the PR.