mergesort: unify ranks loops

mergesort: unify ranks loops

llist_mergesort() has a loop for adding a new element to the ranks array
and another one for rolling up said array into a single sorted list at
the end.  We can merge them, so that adding the last element rolls up
the whole array.  Handle the empty list before the main loop now because
list can't be NULL anymore inside the loop.

The result is shorter code and significantly less object text:

main:
__TEXT	__DATA	__OBJC	others	dec	hex
652	0	0	4651	5303	14b7	mergesort.o

With this patch:
__TEXT	__DATA	__OBJC	others	dec	hex
412	0	0	3441	3853	f0d	mergesort.o

Why is the change so big?  The reduction is amplified by llist_merge()
being inlined both before and after.

Performance stays basically the same:

main:
0071.12: llist_mergesort() unsorted    0.24(0.22+0.01)
0071.14: llist_mergesort() sorted      0.12(0.10+0.01)
0071.16: llist_mergesort() reversed    0.12(0.10+0.01)

Benchmark 1: t/helper/test-tool mergesort test
  Time (mean ± σ):     109.0 ms ±   0.3 ms    [User: 107.4 ms, System: 1.1 ms]
  Range (min … max):   108.7 ms … 109.6 ms    27 runs

With this patch:
0071.12: llist_mergesort() unsorted    0.24(0.22+0.01)
0071.14: llist_mergesort() sorted      0.12(0.10+0.01)
0071.16: llist_mergesort() reversed    0.12(0.10+0.01)

Benchmark 1: t/helper/test-tool mergesort test
  Time (mean ± σ):     109.2 ms ±   0.2 ms    [User: 107.5 ms, System: 1.1 ms]
  Range (min … max):   108.9 ms … 109.6 ms    27 runs

Signed-off-by: René Scharfe <l.s.r@web.de>
Signed-off-by: Junio C Hamano <gitster@pobox.com>