Switch Dashboard to use new gRPC client throughout.

[absoludity]

Description of the change

This PR switches the dashboard over to use the new connect generated grpc-web client and associated generated classes.

It was pretty tedious, with some complexity due to the different generated classes and client.

Non-tedious actual changes

• Updating the way the proto one-of (single choice for different values) work throughout to match the new library (which actually ensures there is only a single value, actually using a OneOf concept, whereas the old improbable library just allowed all the options at once). This was quite confusing to change in the actual code functionality as the code is written as if all options may be specified, but I didn't want to change that in this PR, just switch.
• Updating the shared/ResourceRef class to be a child of the generated ResourceRef class so that we can continue to use it in place of the generated one as we're currently doing (compiler won't accept it otherwise now). Created ResourceRef confusion in operator support #6062 to track removing our shared one if we can.
• Updating the method used in the frontend to stream resources (and changes) for an installed package, since improbable used subscriptions while the connect one uses async iterators (need to verify this works, since I don't think we test it in CI, but it should just fall back to polling if it doesn't).
• Update the KubeappsGRPCClient to use the new client (and the different way of setting headers/metadata).

Tedious parts

• Any object that is generated from the proto messages now needs to be instantiated with, for example, new AvailablePackageSummary({...}) to pass the compilation, rather than the previous {...} as AvailablePackageSummary as this type assertion fails due to the missing methods.
• Updating all the generated class imports with _pb to import the new versions (without deleting the old in this PR, as that blows it out by 16k lines).
• Updating all the enums to the more sensible names used by the new lib
• Update a lot of tests that used jest to spy on the old implementation.
• Update all API method names since the connect one uses snake-case with an initial lower case :/

Benefits

Let's see what CI says, but possibly removing 16k lines of the old generated code (I'll do that in a follow-up PR). The main benefit is that we'll be using a supported grpc-web typescript client and can remove the old improbable generation in the kubeapps apis service.

Possible drawbacks

It's a lot of changes, there maybe unexpected side-effects.

Applicable issues

• ref Replace improbable-eng/grpc-web with grpc/web #6013

Additional information

[netlify]

✅ Deploy Preview for kubeapps-dev canceled.

Name	Link
🔨 Latest commit	a327ed2
🔍 Latest deploy log	https://app.netlify.com/sites/kubeapps-dev/deploys/64100edf5946a40008d5743a

[absoludity]

I've got the image compiling now, but haven't yet sorted out the slightly different behavior for our login checks, which is the cause of the e2e failures.

[absoludity]

I've got the image compiling now, but haven't yet sorted out the slightly different behavior for our login checks, which is the cause of the e2e failures.

Turned out to be a simple difference in the way headers are extracted from a grpc-web response. Locally I checked everything works, including the bits that had changed (long-running requests), let's see what CI thinks.

[absoludity]

I've got the image compiling now, but haven't yet sorted out the slightly different behavior for our login checks, which is the cause of the e2e failures.

Turned out to be a simple difference in the way headers are extracted from a grpc-web response. Locally I checked everything works, including the bits that had changed (long-running requests), let's see what CI thinks.

This is where I miss team-mates to celebrate with :P. After a lot of debugging, found what is hopefully the final CI issue - simply that the incoming error was not being converted correctly because the new grpc error does not have a headerMap like the old. Should have grepped for headerMap much earlier after changing the code in some places.