RFC: Implement codegen command to use AOT-compiled, cached types (Turbo Mode)

[kitten]

Context

Currently, gql.tada is an alternative to codegen tools, which generate types, like the client-preset.

Instead of assigning pre-generated types to GraphQL documents, they're instead parsed and inferred in TypeScript’s type system itself. However, this increases tsc type checking time at least linearly with the amount of documents a user writes and creates.
While this also is a problem with codegen tools (the time of tools run in parallel in a watch mode, or ahead of time, will increase as well), the processing that type inference takes will almost always be much larger.

In #150, we've experimented with printing internal types from TypeScript in the CLI.

Summary

Instead of always inferring the type of GraphQL documents, we could instead allow graphql() to lookup pre-computed types, which are created ahead of time.

This would mean that graphql() has two ways of looking up a type for a given document:

• It first checks whether a cache contains the query its been passed and uses that type if it's available
• If not; it falls back to inferring the type of the given document on the fly

The CLI would be used to populate the cache that graphql() reads from (#76). This is dependent on the CLI having an implementation to discover documents, which is necessary for #77 anyway.

tl;dr: This means that a user can, after they're done editing queries and getting type updates in real-time, run the CLI and commit ahead-of-time-precompiled types for all queries.

Since this is a transparent optimisation, this is like a turbo mode for gql.tada and merges the approaches of codegen and inferred GraphQL types and creates a seamless bridge.

This should also alleviate performance problems when compared to other codegen approaches indefinitely.

Proposed Solution

• gql.tada has a global, declaration-based queries cache with types
• graphql() can lookup types in this cache and use them
• The CLI has a command (name TBD) to generate this cache