Add an immutable rope based textbuffer

[6cdh]

This PR

• Remove the direct framework-backed editor in doclib/editor.rkt.
• Add immutable rope textbuffer and enable for all supported Racket versions.

It also avoids loading framework module, remove the requirement for xvfb-run in headless environment. And improves startup speed and memory footprint.

The treelist based implementation I experimented is still too slow for big files.

[6cdh]

This will be one of the most exciting changes.

• Removed GUI dependency
• Faster textbuffer implementation
• Immutable textbuffer that is very useful for our heavy concurrent use cases
• The new rope-based text buffer implementation can be reused by make a immutable interval map, it will be useful for other services in the future.
• Reduced memory footprint from 400+MB to 200+MB.
• Faster startup time: 1.7s -> 0.9s

[6cdh]

For performance, the benchmark result using bench/run-editor-bench.rkt:

-backend: legacy
-module: /home/lcdh/work/racket-langserver/doclib/textbuffer/editor-legacy.rkt
+backend: rope
+module: /home/lcdh/work/racket-langserver/doclib/textbuffer/editor-rope.rkt
 fixture: lines=2000 width=40 chars=82000

 build editor from source text
 iterations: 50
-cpu: 6368 ms real: 6416.571 ms gc: 715 ms avg-real: 128331.426 us
+cpu: 18 ms real: 18.439 ms gc: 0 ms avg-real: 368.779 us

 copy editor snapshot
 iterations: 100
-cpu: 9424 ms real: 9498.749 ms gc: 1340 ms avg-real: 94987.488 us
+cpu: 0 ms real: 0.093 ms gc: 0 ms avg-real: 0.930 us

 line/char->pos middle
 iterations: 10000
-cpu: 8 ms real: 8.677 ms gc: 0 ms avg-real: 0.868 us
+cpu: 2 ms real: 2.803 ms gc: 0 ms avg-real: 0.280 us

 pos->line/char middle
 iterations: 10000
-cpu: 13 ms real: 13.161 ms gc: 0 ms avg-real: 1.316 us
+cpu: 2 ms real: 2.810 ms gc: 0 ms avg-real: 0.281 us

 get-char middle
 iterations: 100000
-cpu: 112 ms real: 113.493 ms gc: 0 ms avg-real: 1.135 us
+cpu: 12 ms real: 12.887 ms gc: 0 ms avg-real: 0.129 us

 get-text full document
 iterations: 200
-cpu: 658 ms real: 663.656 ms gc: 4 ms avg-real: 3318.281 us
+cpu: 102 ms real: 103.083 ms gc: 1 ms avg-real: 515.417 us

 get-text middle range
 iterations: 5000
-cpu: 20 ms real: 20.099 ms gc: 0 ms avg-real: 4.020 us
+cpu: 5 ms real: 5.921 ms gc: 0 ms avg-real: 1.184 us

 single top-of-file insert with restore
 iterations: 2000
-cpu: 228 ms real: 230.751 ms gc: 0 ms avg-real: 115.375 us
+cpu: 7 ms real: 7.542 ms gc: 0 ms avg-real: 3.771 us

 single middle replacement with restore
 iterations: 2000
-cpu: 501 ms real: 505.195 ms gc: 0 ms avg-real: 252.597 us
+cpu: 20 ms real: 21.138 ms gc: 0 ms avg-real: 10.569 us

 small multiline replace with restore
 iterations: 1000
-cpu: 343 ms real: 346.685 ms gc: 1 ms avg-real: 346.685 us
+cpu: 10 ms real: 10.841 ms gc: 0 ms avg-real: 10.841 us

 typing burst of 50 inserts with restore
 iterations: 100
-cpu: 363 ms real: 366.392 ms gc: 0 ms avg-real: 3663.916 us
+cpu: 16 ms real: 16.392 ms gc: 0 ms avg-real: 163.923 us

[6cdh]

In CI, removing the GUI dependency and adding the -j flag to raco test cut test runtime from 50s to 14s. It's 7s on my machine. This will make iteration significantly faster.