Tweaks in type stubs generation

[contagon]

Ever since stub generation was built into pyscript, I've been leaning heavily into python typing (using pyrefly as my typechecker currently, previously basedpyright) as I find this has been catching A LOT of the bugs I had in my automations.

There was always a couple of typing oddities the type checkers didn't like in the pyscript stubs. I've been fixing them by hand for a while, but finally took the time to fix them in the generation.

DISCLOSURE: A lot of the code in the PR was LLM generated, but I reviewed it to the best of my ability. Also the PR is all hand-written:)

Missing entities

I had a number of missing entities in the stubs, namely zone.home and a number of template entities were the ones I noticed. This PR now uses the state machine (sorted to get more deterministic output) instead of the entity registry to get ALL entities. From what I can tell, this should include everything the entity registry also did.

Poor defaults

There was a number of defaults in the stubs that didn't type check correctly. Examples include delay_secs: int = 0.4, activity: int = None, and pipeline : str = 1. These are now transformed into delay_secs: float = 0.4 , arg: int | None = None, and pipeline: str | int = 1.

Tests & Type Checkers

Tests have been added for each of these! I'd also be happy to add a section to the docs about setting up a type checker with pyscript. Between stubs giving autocompletion and the type checker making sure I call services correctly (bad types when calling -> type errors) and am only using entities that still exist (deleted entities -> generate new stubs -> type errors anywhere they are used), writing pyscript now feels just like native python code!

I'd recommend giving this a shot before merging, just to make sure it all works! Also happy to tweak whatever