Attempt to optimize charmaps

[Rangi42]

This came up in discussion regarding https://burntsushi.net/transducers/.

There are a lot of possible ways we could optimize our charmap data structure and algorithms. So far this PR only implements one, and hasn't tried measuring whether it's even a reliable improvement yet.

[Rangi42]

Tested this on data/maps/map_data.asm from polishedcrystal, which has a lot of the game's text. Comparing 5 runs of time rgbasm -Weverything -Wtruncation=1 -E -Q8 -P includes.asm -DDEBUG -o data/maps/map_data.o data/maps/map_data.asm with a truncated mean (mean after discarding min and max):

• current: [26.599s, 26.669s, 28.228s, 28.877s, 29.631s] = avg 27.925s
• this PR: [26.492s, 26.502s, 26.619s, 27.149s, 27.834s] = avg 26.757s (1.04x speedup)

Not sure if this is significant, but it's at least not harming performance so far.

Also, to measure maximum RAM usage, /usr/bin/time -v rgbasm ... reports the "Maximum resident set size (kbytes)":

• current: [34700, 34704, 34704, 34708, 34728] = avg 34705 KB
• this PR: [32480, 32480, 32480, 32480, 32480] = avg 32480 KB (1.07x improvement)

[ISSOtm]

This PR looks no worse than the previous implementation, but I'm thinking that this entire module is starved for comments, heh.