Added new base path and cleanup logic to fix tasty determinism

[Garrett-Thomas]

What:

• sourceJars were extracted into random tmp directories. The tasty files then contained these random file-paths in their metadata.
• The final jar that was produced was non deterministic because of this behavior.
• Source extraction now happens in the parent directory of the output jar path in a folder /tmp. e.g. /bazel-out/x86/target/tmp/<source>.
• That directory is checked and deleted before building. The main logic in work is wrapped inside a try finally so that the folder is always deleted.

Why:

• By removing this source of non-determinism, we increase the reproducibility and the cache-ability of our codebase.

Testing:

• There was already a non-determinism test, but it tested only on a single .scala file. I zipped this file into a sourceJar, making it so that source extraction happens before compilation. The test already performs a byte-for-byte comparison on the compiled output.

[Garrett-Thomas]

Formatter

[jjudd]

Thanks for doing this and for making the directory deletion more robust. I spent some time poking at this PR and it LGTM.

I'll merge this and include it in the release I intend to cut tomorrow after merging my PR. I'll get this to the monorepo tomorrow with the PR I've got open over there.

[jadenPete]

Thanks for doing this!

I probably would've left a more detailed comment explaining why the temporary directory needs to be where it is to prevent the bug we were seeing, but otherwise, this looks good to me!