Vendor tinyjson sources to make process_wrapper deterministic

[jfirebaugh]

Fixes #1530

[scentini]

I was also working on fixing this today over at #1728. I believe it's the more desireable approach as it will allow us to to add further dependencies of the process wrapper (should we need them) without vendoring them.

[jfirebaugh]

I saw that! Either approach seems viable to me. I'm happy with whatever the maintainers decide.

[UebelAndre]

I too would lean away from vendoring code if possible 😄

[jfirebaugh]

The approach in #1728 would prevent fixing #1601.

[jfirebaugh]

In main...jfirebaugh:rules_rust:jfirebaugh/integration_tests, I wrote an integration test for the issue using rules_bazel_integration_test. It fails prior to this fix and passes afterward.

Let me know what you think of that approach. It could be useful no matter which of the two PRs gets merged, and a good basis for writing integration tests in general.

[UebelAndre]

The approach in #1728 would prevent fixing #1601.

I'm generally not a fan of vendoring code to projects like this but tinyjson is indeed tiny. I'm curious how exactly this solves determinism issues though. Is it deterministic based on where you checkout the repo or otherwise?

Additionally, does it make a difference if tinyjson is it's own target vs a part of the process_wrapper itself? I would prefer it if tinyjson could remain it's own package. Finally, could a tool (small script) be added to re-vendor the project? I would like make bringing in new changes as painless as possible.