Cache generated matlab references

[faridyagubbayli]

closes #259

Adds caching support for the generated matlab references. This will allow us to avoid re-generating the reference files in every test run -- speed up our tests, improve developer experience and reduce Github's energy bill :)

Introduced logic is simple: if there is no change in the matlab reference generation files, pull previously generated files (if they exist). I also changing data sharing logic between Github Action-Jobs. Previously, we used artifacts to share the data. Now we are using the cache directly. Cache functionality is faster compared to artifacts from what I observed.

Some numbers:

	Matlab reference generation duration	Total test duration
Artifacts based workflow (without caching)	3 mins 36 secs	8 mins 9 secs
Caching based workflow (cache-miss)	2 mins 22 secs	6 mins 53 secs
Caching based workflow (cache-hit)	15 secs	5 mins 20 secs

^ here first and second scenarios are similar because we generate the reference files in both of them. Latter is faster due to usage of cache to share data between jobs.
^^ when there is cache-hit, pytest runs start significantly quicker than first two scenarios. This means you can start reading test logs faster than before.

Future improvement: Use file-level caching. Currently, if any of the matlab reference generation files get updated, all references will be re-generated. In the future, we can work towards having more fine-grained caching.

[waltsims]

Please see #264 for my suggested changes.

	Matlab reference generation duration	Total test duration
Artifacts based workflow (without caching)	3 mins 36 secs	8 mins 9 secs
Caching based workflow (cache-miss)	2 mins 22 secs	6 mins 53 secs
Caching based workflow (cache-hit)	15 secs	5 mins 20 secs
New Caching based workflow (cache-miss)	2m 45s	4m 23s
New Caching based workflow (cache-hit)	6s	4m 6s

I don't know how indicative these numbers are, but the workflow file is simpler.

[faridyagubbayli]

I don't know how indicative these numbers are, but the workflow file is simpler.

Agree, they are not proper performance benchmarks. I just wanted to demonstrate the speed up based on the reported run times in Github. I used only a single test run for each row.