V2: Update query keys for partial matching

[timostamm]

This changes the type for Connect query keys to be:

type ConnectQueryKey = [
  /**
   * To distinguish Connect query keys from other query keys, they always start with the string "connect-query".
   */
  "connect-query",
  {
    /**
     * A key for a Transport reference, created with createTransportKey().
     */
    transport?: string;
    /**
     * The name of the service, e.g. connectrpc.eliza.v1.ElizaService
     */
    serviceName: string;
    /**
     * The name of the method, e.g. Say.
     */
    methodName?: string;
    /**
     * A key for the request message, created with createMessageKey(),
     * or "skipped".
     */
    input?: Record<string, unknown> | "skipped";
    /**
     * Whether this is an infinite query, or a regular one.
     */
    cardinality?: "infinite" | "finite";
  },
];

The function createConnectQueryKey is updated to accept key parameters as an options object:

createConnectQueryKey({
  transport: myTransport,
  schema: ElizaService.method.say,
  input: create(SayRequestSchema, { sentence: "hi" }),
});

The transport is now part of the key to support multiple transport without tainting the cache (#418).

All key parameters except schema are optional, which allows to create partial keys (#375) for filtering.

The updated function replaces createConnectInfiniteQueryKey. To create a key for an infinite query, use cardinality: "infinite" and the pageParamKey option.

[timostamm]

I missed that the "page param" is not typically part of the key for infinite queries. Seems like a good fit to support it here. It means that the users doesn't have to omit the page param field with createConnectQueryKey.

[timostamm]

The expression needs to move up. Otherwise, a new transport is created every time the hook is rendered, and the key changes.

As noted in #418 (comment), this can be a footgun. I think it should be sufficient to call it out in the documentation, and to give an example. WDYT, @paul-sachs?

[paul-sachs]

Agreed. Once we've taken transport into account as a key, it's important that it gets memoized

[paul-sachs]

Nice cleanup!

[paul-sachs]

Hmm, interesting. I guess this allows for invalidating a whole service if necessary.