Improve performance of location data compensation

[helixbass]

Fixes #5290

As discussed in that issue, performance for files with lots of "location data compensations" (eg stripped carriage returns, so all files on Windows) was blowing up because the algorithm used for calculating location data compensation was very inefficient

So this PR improves that algorithm to one that shouldn't have significant performance implications

To test, I made a copy of a decent-size source file eg src/lexer.coffee and converted it to have Windows-style line endings (\r\n), and ran bin/coffee -b -p [converted-filename.coffee]

Before this fix, compilation was taking many seconds, and with this fix, it took less than a second

I haven't done specific performance comparisons of a file with Windows line endings when you omit the location data compensation calculation entirely vs with this new algorithm, that might be interesting as a baseline

[helixbass]

This just changes the iteration when calculating the location data compensations for a given start/end range to iterate on start..end instead of iterating on the entire @locationDataCompensations data structure every time

[GeoffreyBooth]

So you don't need Windows to test this, just a file with Windows line endings. To make things a bit more painful, I created a copy of nodes.coffee with Windows endings:

 ✦ time cat src/nodes.coffee | coffee --stdio --compile > /dev/null
cat src/nodes.coffee  0.00s user 0.00s system 4% cpu 0.081 total
coffee --stdio --compile > /dev/null  2.21s user 0.17s system 158% cpu 1.500 total

 ✦ perl -p -e 's/\n/\r\n/' < src/nodes.coffee > test.coffee # https://superuser.com/a/71509/138751

 ✦ time cat test.coffee | coffee --stdio --compile > /dev/null
cat test.coffee  0.00s user 0.00s system 4% cpu 0.083 total
coffee --stdio --compile > /dev/null  248.14s user 1.87s system 102% cpu 4:04.77 total

 ✦ git checkout --track helixbass/iss5290-slow-location-data-compensation
Branch 'iss5290-slow-location-data-compensation' set up to track remote branch 'iss5290-slow-location-data-compensation' from 'helixbass'.
Switched to a new branch 'iss5290-slow-location-data-compensation'

 ✦ time cat test.coffee | ./bin/coffee --stdio --compile > /dev/null
cat test.coffee  0.00s user 0.00s system 2% cpu 0.127 total
./bin/coffee --stdio --compile > /dev/null  2.21s user 0.17s system 149% cpu 1.585 total

So in this test, compiling a version of nodes.coffee with Windows line endings dropped from four minutes in 2.5.0 to 1.5 seconds using this branch, the same amount of time it took to compile the Unix line endings version. Seems like a success to me.