Support record/replay for supportability and testing scenarios

[lukehoban]

Although users interface with Pulumi through a program written in a particular programming language, the behaviour of that program ultimately boils down to a sequence of RPC calls made to the Pulumi engine. For many use cases, it is useful to be able to record, share, and reply these RPC calls:

1. For supportability scenarios, such as debugging a user problem without requiring them to share source code.
2. For testing scenarios, such as being able to lock in, in a language agnostic way, the expected results of a program or component.

For testing, we have a few types of tests that are used frequently in Pulumi today:

1. Integration tests, which run a program written in a supported Pulumi programming language against a real cloud provider
2. Unit tests, which run within the program as part of the language's built-in unit testing capability, using mocks instead of a real cloud provider.

There are cases though where we would like to test in a more language-agnostic way, and decouple testing between the language/component layer (which generates a sequence of RPC calls) and the engine (which handles those calls).

It would be nice if we had a language-agnostic framework to run Pulumi programs against a mock monitor RPC server, where
we would could supply the canned responses to particular RPC requests through config.

As a bonus, it would be nice if we could capture the RPC request/response when running a Pulumi program normally to use
that as config to run subsequent tests with Pulumi programs written in another language.

Some explorations of these ideas were done in #11710 and #11085.