Support worker multiplexing and multiplex sandboxing

[jadenPete]

This issue is part of a broader proposal to merge some of the features in lucidsoftware/rules_scala into this ruleset.

The tracking issue for this effort

The relevant proposal section.

From the documentation on multiplex workers:

Multiplex workers allow Bazel to handle multiple requests with a single worker process. For multi-threaded workers, Bazel can use fewer resources to achieve the same, or better performance. For example, instead of having one worker process per worker, Bazel can have four multiplexed workers talking to the same worker process, which can then handle requests in parallel. For languages like Java and Scala, this saves JVM warm-up time and JIT compilation time, and in general it allows using one shared cache between all workers of the same type.

Although multiplex workers are an experimental feature, we've been using them in lucidsoftware/rules_scala for years to great effect and believe the mainline ruleset would benefit significantly from them.