feat(backend): clean up choose_precision

[psychedelicious]

Summary

• Allow user-defined precision on MPS.
• Use more explicit logic to handle all possible cases.
• Add comments.
• Remove the app_config args (they were effectively unused, just get the config using the singleton getter util)

Related Issues / Discussions

https://discord.com/channels/1020123559063990373/1049495067846524939/1226123179303370794

QA Instructions

Test bfloat16 precision on MPS on torch 2.3 nightly @Vargol

Pretty simple change.

Merge Plan

n/a

Checklist

• The PR has a short but descriptive title, suitable for a changelog
• Tests added / updated (if applicable) n/a
• Documentation added / updated (if applicable)

[Vargol]

TL;DR - seems to be working just fine.

Long version...
Okay, updated my env using main with pull request

M3iMac:InvokeAI $ git fetch origin pull/6171/head:test_6171
From https://github.com/invoke-ai/InvokeAI
 * [new ref]             refs/pull/6171/head -> test_6171
M3iMac:InvokeAI $ git switch test_6171
Switched to branch 'test_6171'

build the front end and pip uninstalled invokeai and re-installed using the repo

Ran on release torch, expecting a BFlaot16 not supported error, mostly as a way to check it was using bf16

[2024-04-07 12:13:56,513]::[InvokeAI]::ERROR --> Error while invoking session b8a518bb-5b8c-45a0-b9aa-2c8fd4eb71e5, invocation c61d4e77-a3e9-41e6-b801-6884ff70a366 (sdxl_compel_prompt):
BFloat16 is not supported on MPS

So good so far.
Updated torch to 2.30 release candidate

(InvokeAI) M3iMac:InvokeAI $ pip uninstall torch torchvision
...
  Successfully uninstalled torch-2.2.1
... 
 Successfully uninstalled torchvision-0.17.1

(InvokeAI) M3iMac:InvokeAI $ pip3 install --pre torch torchvision --index-url   https://download.pytorch.org/whl/test/cpu
...
Successfully installed torch-2.3.0 torchvision-0.18.0

Restart invokeAI and ran a few 'quick' prompts including a lora and a controlnet which ran to completion.
The simple ones worked fine I may have , I may have hit a bug in LoRA's but that seems to happen in float16 too.
So I don't think its related.