Improve cigarstring performance

[limwz01]

Use direct string manipulation instead of inefficient Python joins

[limwz01]

@AndreasHeger @jmarshall @bioinformed any chance of merging this? Thanks!

[jmarshall]

This PR is certainly on my shortlist to merge in due course, though I'd probably prefer if it used kstring rather than manual memory management. Have you done any timing testing?

[limwz01]

I did timing testing quite long ago, and there was substantial improvement, but I've just tested with the latest version on Python 3.12.1 using a ~100MB BAM file:

    tm = timeit.default_timer()
    with pysam.AlignmentFile(args.bam) as in_bam_f:
        for read in in_bam_f:
            read.cigarstring
    print("finished reading BAM in %f seconds"%(timeit.default_timer() - tm))

without optimisation: 5.821680s
with optimisation: 3.519799s

[limwz01]

nice, kstring is even faster even though I don't see it using a pre-allocated array:

without optimisation: 5.821680s
with optimisation: 2.439370s

There was a problem using the pattern in to_string:

            ret = force_str(buf.s[:buf.l])

            if buf.m:
                free(buf.s)

            return ret

I had to revert to using try:

            try:
                return buf.s[:buf.l].decode("utf8")
            finally:
                if buf.m:
                    free(buf.s)

If anyone can help to make the original pattern work, that would be great!