refactor[venom]: make venom repr parseable

[charles-cooper]

What I did

How I did it

How to verify it

Commit message

make the `__repr__()` implementations for venom data structures
(`IRContext`, `IRFunction, `IRBasicBlock`) emit strings which will
round-trip through the parser.

for labels generated in the frontend which are not necessarily
valid identifiers (e.g. `"internal 5 foo()"`), these are represented as
escaped strings. the expedient way to implement this was to simply use
`json.loads` / `json.dumps`; there did not seem to be any convenient
stdlib or lark function to do this. since this adds grammar complexity,
the other method that was considered was to map all labels to valid
identifiers in `ir_node_to_venom.py`. but this approach seems easier
than cleaning up all the non-identifier labels generated by the
frontend; plus being able to have arbitrary strings in labels seems like
it will come in handy during debugging some time.

a couple other grammar updates/fixes:
- update instruction order in the text format for `phi` and `invoke`
- ensure instructions are terminated with newline (otherwise they can
  continue slurping tokens from the next line).
- allow signed ints inside `CONST` nodes as `IRLiteral` accepts negative
  numbers

misc/refactor:
- remove a dead function (`str_short()`).
- remove a dead branch in `ir_node_to_venom.py`
- when optimization level is set to `CODESIZE`, the roundtrip test
  is set to xfail, as the data section contains bytestrings (which do
  not parse yet).

Description for the changelog

Cute Animal Picture

[HodanPlodky]

Just wondering if it would be better to just change up some of the labels instead of parsing escaped string (I dont know how many cases are there I stumbled so far on case with the space which could be replaced by underscore)?

[charles-cooper]

i could do it both ways honestly, but this seemed nicer for debugging (you can imagine the frontend generating a sentence or description of where it came from for a label)