Fix decompress_peek_gz to cope with files starting on empty gzip blocks.

[jkbonfield]

Demonstration:

$ echo -e "@seq\nAAAAA\n+\nABCDE" | gzip > _normal.gz                           
$ (gzip < /dev/null;cat _normal.gz) > _empty.gz                                 
                                                                                
$ ./htsfile _normal.gz                                                          
_normal.gz:     FASTQ gzip-compressed sequence data                             
                                                                                
$ ./htsfile _empty.gz                                                           
_empty.gz:      empty gzip-compressed data                                      
$ ./test/test_view _empty.gz                                                    
Unsupported or unknown category of data in input file                           

Zlib's inflate ends with Z_STREAM_END and then repeated calls to get more won't give any difference, so we need the explicit reset to move forwards with the next block.

[jkbonfield]

Now with added paranoia for loop detection. Just incase...

[jmarshall]

If total_out may have been reset by inflateReset, is it still appropriate to use it in the loop above?

[jkbonfield]

Doh! Good point. No (although in practice avail_in seems to catch it).

I'll make a follow up fix with the similar next_out - dest logic instead.

[jkbonfield]

It appears I was being overly paranoid (although I left in the loop detection bit, even though I can't actually trigger zlib to consume no input without producing an error).

avail_out would have been the correct variable, as unlike total_out this isn't reset. However it turns out neither are necessary. Calling inflate with one of those as zero returns an error (Z_BUF_ERROR) so it promptly bails out the loop anyway, and even if it just returned Z_STREAM_END again it'd be caught by the other checks.

So it's now much simpler. See #1646 for 2nd try at fixing this :)