Performance and optimization improvements, changes to encoding

[Rob23oba]

This pr replaces encoding.js with a more optimized version which works with a wasm module (or arraybuffer). Some formats were changed to accomodate this, e.g.: immediates are no longer encoded, e.g. [ Opcodes.i32_const, ...signedLEB128(x) ] -> [ Opcodes.i32_const, x ] which makes a lot of reading much more straightforward, the entire opcode is always in the first index, e.g. Opcodes.i32_trunc_sat_f64_s -> [ Opcodes.i32_trunc_sat_f64_s ]. Additionally, the backtrace system has been reworked to use binary search to find the correct instruction instead of searching everything. Because of the previous two changes, a lot of other code needed to be reworked in order to fit the new scheme. Also: block types in Porffor.wasm are now encoded as their name instead of their id and multi-value block types are also supported (e.g. [ Opcodes.if, [ [ Valtype.f64, Valtype.i32 ], [ Valtype.i32 ] ] ] which is encoded if f64 i32 -> i32 in Porffor.wasm). Because it bothered me there is also now a more proper way of doing constant evaluation with opt.js (giant table jumpscare) and optimizations relating to drop have been generalized (oh wow this so much more involved than before). (yeah I couldn't resist adding optimizations)