Consider using a real language parser to parse MI output

**Describe the bug**
Currently, we use regular expressions to parse the GDB MI output. In https://github.com/cs01/pygdbmi/pull/72#pullrequestreview-1066128398 Chad suggested we used a proper parser to do it:

> NIce, this is much better!
>
> Should we consider writing a grammar to simplify this more? Someone did this in rust for pygdbmi -- https://github.com/JP3BGY/gdbmi/blob/master/src/gdbmi_output.pest
>
> We could use lark which has implementation in other languages for the same source `.lark` grammar.
>
> * https://github.com/lark-parser/lark
> * https://lark-parser.readthedocs.io/en/latest/json_tutorial.html
> * https://www.lark-parser.org/ide/#
> * https://www.lark-parser.org/Lark.js/index.html

I'm not keen to write a grammar (it's not something I've done since univeristy!) but, as we already have one we could use, this is a good idea.

**To Reproduce**
N/A

**Expected behavior**
No visible change to users (apart from performance improvements probably).

**Screenshots**
N/A

**Please complete the following information:**
* OS: All
* pygdbmi version (`pip freeze` output): `master` pre-0.10.0.3.

**Additional context**
N/A

Start by locating the regular-expression code that parses GDB MI output and compare its behavior with the referenced Rust grammar and Lark documentation. Done means replacing the current parsing approach without visible user-facing changes, while preserving existing output handling and checking for the expected performance improvements.

Written by the indexing model from the issue text.