Don't copy ~every input string if we can avoid it.

[kivikakk]

Turns out process_line will pretty much duplicate every line fed to it — we avoid including the actual line ending character in feed, but then process_line demands its input ends in one!

This results in a ~1.12x speed-up on the benchmark document. Yippee. Documents with CRLF line actually already ran about as fast — they get a more modest 1.05x speed-up.

[github-actions]

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
./bench.sh ./comrak-38ebabf	113.5 ± 0.9	112.0	115.8	1.99 ± 0.03
./bench.sh ./comrak-main	128.2 ± 10.8	124.5	180.0	2.25 ± 0.19
./bench.sh ./pulldown-cmark	57.0 ± 0.8	56.1	59.5	1.00
./bench.sh ./cmark-gfm	86.9 ± 8.1	80.3	104.0	1.52 ± 0.14
./bench.sh ./markdown-it	250.3 ± 2.4	244.5	256.5	4.39 ± 0.07

Run on Sun Oct 19 02:11:47 UTC 2025