Prune chain candidates that cannot start a chain

[ksahlin]

extract_chains sorts all anchors with a score of at least 0.7 times the
best score (the candidates) and then starts a chain from each candidate
that is not yet part of another chain. An anchor j that is the
predecessor of a candidate i with a strictly higher score can never
start a chain: i comes before j in the sorted list, and when it is i's
turn, either i is unused and its walk visits j first, or i was already
marked as used by a walk that continued to j.

Such anchors are now removed from the candidates before sorting. They
are marked in the loop that collects the candidates (by setting their
score to -inf, which cannot be a real score), so no extra pass over the
anchors is needed. Output is unchanged.

Speedup on an Apple M4 Max, -x, 1 thread, median of 5 runs:

• maize, 50 nt reads: chaining -3% to -4.5%, total mapping time -3%
• maize, 150 nt reads: chaining -1.5% to -2%, total mapping time -1%
• rye, 50 nt reads: chaining -3%
• chrY, CHM13, rye 150 nt and all 10 kb reads: unchanged

If the speedup is very marginal also when evaluated independently,
it might not be worth it. Let's discuss.