Implement `debugPanic` like functionality for `getConfig` and `getDmmf` of Query Engine

To be able to implement and test Prisma’s CLI engine crash reporting for getConfig and getDmmf, we need new functionality similar to debugPanic of Migration and Introspection Engines (https://prisma.github.io/prisma-engines/doc/migration_core/json_rpc/index.html#-debugpanic) that we can somehow trigger when we execute getConfig or getDmmf.

For getConfig we call in the CLI:

• For the Node-API library NodeAPIQueryEngineLibrary.getConfig(...)
• For the Binary ...engineArgs, 'cli', 'get-config', ...args

For getDmmf we call in the CLI:

• For the Node-API library NodeAPIQueryEngineLibrary.dmmf(datamodel)
• For the Binary '--enable-raw-queries', 'cli', 'dmmf'

For all 4 we have 2 different options how to implement this:

1. We implement a separate function/endpoint debugPanic that we can call instead of the intended function/endpoint, and when we want to test the error reporting we would just call out to that instead of the actual one.
2. We use the existing function/endpoint but add a way to configure what happens inside (and then trigger the panic there). For Node-API library that could maybe be an environment variable, and for the Binary an additional parameter.

[janpio]

That sounds good. @jkomyno who will use these, just needs to know how to trigger each.

What you describe for Node-API could be related to these issues (and the comments): https://github.com/prisma/prisma/issues/7810 + https://github.com/prisma/prisma/issues/11362

[dpetrick]

The binary already has a way of triggering errors: Fatal or non-fatal, based on included headers x-debug-non-fatal or x-debug-fatal. Fatal = engine exists, non-fatal: panic is thrown. That’s only possible on the graphql endpoint right now, but not really a problem to expand that to other ones too.

The NodeAPI already has a debug_panic endpoint that simply errors.

Edit: The NodeAPI basically returns a user-facing error, not really panics. We could change that to a true panic, but then the assumption is that there are cases where the panic hooks are not working. The NodeAPI should always hook panics with the handler, but that requires some more investigation.