Update caches systems references to be actual caches and not an additional 0(n) lookup

[SimonDarksideJ]

Overview

The Service registry is effectively a 0n lookup system using a dictionary, the dictionary is used primarily to ensure fast 0n lookup and also provide unique adherence to the single service principle of dependency injection.

This change resolves an issue whereby the current "CoreServices" systems lookups are actually not caching / improving lookup performance by adding an additional 0n lookup on top of / in replacement of the original 0n lookup.

Simply put, this provides direct references to the the instantiated services in the service registry rather than looking them up, potentially every update frame.

I have left in the updated cache table, although it is likely redundant.

Changes

Simply adds direct reference caches for each of the core systems identified in the "Core Systems" cache.

Verification

This optional section is a place where you can detail the specific type of verification
you want from reviewers. For example, if you want reviewers to checkout the PR locally
and validate the functionality of specific scenarios, provide instructions
on the specific scenarios and what you want verified.

If there are specific areas of concern or question feel free to highlight them here so
that reviewers can watch out for those issues.

As a reviewer, it is possible to check out this change locally by using the following
commands (substituting {PR_ID} with the ID of this pull request):

git fetch origin pull/{PR_ID}/head:name_of_local_branch

git checkout name_of_local_branch

[keveleigh]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 2 pipeline(s).

[keveleigh]

We investigated this approach a while back (CC @MaxWang-MS) and decided against it, but I don't remember the exact reason now...

If we go this approach now, we'll probably want to clear these new caches in ResetCacheReference and ResetCacheReferences

[SimonDarksideJ]

I'll look to update it and clear the caches from that call. Although I am concerned why you would even need to reset the caches as all the systems listed in this function are transient, meaning they should never be cleared for the lifetime of the runtime.

[SimonDarksideJ]

Updated as per your request @keveleigh

[keveleigh]

all the systems listed in this function are transient, meaning they should never be cleared for the lifetime of the runtime.

Just to make sure we're talking about the same thing, my understanding of "transient" and "should never be cleared for the lifetime of the runtime" are two different things: the first is short-lived, and the other sounds more long-lived.

Either way though, we don't assert that MRTK services are specifically one or the other! Devs may swap profiles at runtime, leading to these cached references potentially going stale. The cache clearing logic all happens in MixedRealityToolkit.cs when the profile is changed and the services are torn down and spun back up.

[keveleigh]

I'm also going to investigate cherry-picking this to 2.8.1. I think this will help with a selection of perf issues I've been strategically targeting. Thanks!

[keveleigh]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[keveleigh]

Thanks for this change!