gateway

Currently, UnframedGrpcService returns an error message as plain text.
https://github.com/line/armeria/blob/54b70906bcd813f0b9b2c6715ad4086160ac49b2/grpc/src/main/java/com/linecorp/armeria/server/grpc/UnframedGrpcService.java#L255-L272
It would be nice to return a JSON object if the content type is JSON compatible.

{ 
   "code": "FAILED_PRECONDITION",
   "message": "Input validatio

Is your feature request related to a problem? Please describe.

As observed during #261, users can accidentally set timeouts inconsistently. A concrete example is that one can set a User facing timeout lower than any backend timeout, such that requests may fail early.

Describe the solution you'd like

We should provide some sanity checking to prevent unintended timeout behavior.

It looks like the Mercurius Gateway service expects "errors" to exist in the response, however, if I have a fastify plugin that throws an error before it ever hits GraphQL, then the gateway emits an error.

{
  "errors": [
    {
      "message": "Cannot read property 'profile' of undefined",
      "locations": [
        {
          "line": 2,
          "column": 3
        }
   

Is there a plan to build kernelspec with pytorch support? If not, what's the right way to contribute?
The contribution instruction page seems to be not available: https://jupyter.readthedocs.io/en/latest/contributor/content-contributor.html