Cache SLAM Dataset for CI

[mhuzai]

EuRoC V1_02_Medium gets downloaded and unzipped in every single test run. This not only makes tests run longer, but probably also creates (a lot of) unnecessary traffic for the ETH server. And this is about to get worse because we just added a bunch of new CI pipelines. So, we need to find a way to download and unzip this dataset once, cache it, and then use the cache in the test runs.

[charmoniumQ]

Clarification: Every single test run in CI. Local tests (with a persistent disk) do not suffer this problem.

We need to use GitHub actions caching here.

[e3m3]

And this is about to get worse because we just added a bunch of new CI pipelines.

Each CI pipeline (flow in ILLIXR/configs/ci/flows introduced in #270) is still embedded within one call to Runner at the moment, so the download is amortized over each pipeline. However, ideally each pipeline will be processed independently by CI to achieve parallelism, necessitating caching.

[charmoniumQ]

The dataset gets downloaded to .cache. Saving it also saves the external plugins (Kimera).

We should also consider moving .deps into .cache. Same with the conda env.

[e3m3]

We should also consider moving .deps into .cache. Same with the conda env.

These are now in .cache in #264.