Sourcemap performance improvements (#668)

This PR brings two performance improvements to the `rojo sourcemap`
command:

- Use `rayon` with a small threadpool to parallelize sourcemap
generation while still keeping startup cost very low
- Remove conversions to owned strings and use lifetimes tied to the dom
instead, which mostly improves performance with the
`--include-non-scripts` flag enabled

From my personal testing on an M1 mac this decreases the sourcemap
generation time of our games by 2x or more, from ~20ms to ~8ms on one
project and ~30ms to ~15ms on another. Generation is pretty fast to
begin with but since sourcemaps are heavily used in interactive tools
(like luau-lsp) a difference of a couple frames can be great for ux.

Cargo.toml

@@ -72,7 +72,9 @@ jod-thread = "0.1.2"
 log = "0.4.14"
 maplit = "1.0.2"
 notify = "4.0.17"
+num_cpus = "1.15.0"
 opener = "0.5.0"
+rayon = "1.7.0"
 reqwest = { version = "0.11.10", features = ["blocking", "json", "native-tls-vendored"] }
 ritz = "0.1.0"
 roblox_install = "1.0.0"