0.8.3

Bali 0.8.3 is out after quite a long time. I'm mostly just preparing it to be as stable and usable for a project. It comes with about ~40 commits.

Fixes

Passes in Madhyasthal are now a sequence, not a set. This allows certain passes to be repeated after certain other passes.
Madhyasthal's copy propagation pass is more aggressive now
Madhyasthal's DCE pass is more careful when dealing with globals now
The tokenizer is now 100% pure and does not use logging as there was no point in doing so.
The parser no longer uses logging as there was no point in doing so.
Bytecode generation no longer crashes in certain cases if a reassignment's literal is a null.
bali::grammar::Tokenizer is no longer a GC'd type as it doesn't require it.
The parser no longer (ab)uses deepCopy() for the tokenizer.
The x64 midtier JIT now uses more diagnosable constants instead of random values that might cause undiagnosable crashes in edge cases (dummy jump addresses).
String.prototype.indexOf() now relies solely on strutils::find(), as Kaleidoscope was way too finnicky and unpredictable in many cases, while providing a negligible performance improvement.
Fixed a small bug in Tokenizer::advance() that'd make column reporting inaccurate in some cases

Massive speedups in the bytecode interpreter by no longer relying on a function to find the currently executing clause/function. This was incredibly hot and required way too much computation for an otherwise simple task.
Slimmed down bali::runtime::vm::interpreter::Operation by removing old fields from very old experiments.

Madhyasthal can now lower and compile the following operations, allowing it to compile a lot of loops:
* LTI
* INC
* GTE
* GTEI
Madhyasthal's x64 can now perform proper conditional jump patching. It just takes the preexisting jump patching code to add some extra metadata required for emitting conditional jumps.
Laid down a lot of infrastructure for mid-execution JIT patching, letting compiled segments take over from the middle of interpreting a function, possibly providing massive performance speedups. Currently disabled as it caused major semantic breakage, but I'd love to revisit it in the future.
Bumped nim-url to 0.2.0 from 0.1.0, this brings various fixes and major performance optimizations along with it, along with AArch64 SIMD support for its parser.