feat: schema codegen bucketing to help with tree-shaking

[kuhe]

Context

In the implementation of Schema serde #1600, the schema objects, corresponding roughly 1-1 with modeled shapes, expand linearly with the model size.

Initially, all schemas were generated into a single file called /schema/schemas.ts. Because schema objects call factory functions, they are not tree shaken correctly. Even adding @__PURE__ etc. annotations in the source code did not work, with both webpack and esbuild failing to tree shake correctly (rollup did fine).

PR

Here we have a bucketing system to generate schema objects into separate files.

• Generating a single file per schema object would be excessive: this creates too many files, which is destructive to our package size measurements and to the file-&-build-systems that would be handling our packages.
• The question was then, how do we logically bucket schemas into separate files?

Solution / heuristic

We can't arbitrarily put e.g. 100 schemas per file into multiple bucketing files. Schema objects form a dependency tree, and random bucketing would create poor tree-shaking outcomes.

The solution in the PR buckets schema objects into a logical operation group. For example, in the AWS S3 model, the following schema objects are grouped together in a file called schemas_Payment.ts

export var GetBucketRequestPaymentOutput = struct()
export var GetBucketRequestPaymentRequest = struct()
export var PutBucketRequestPaymentRequest = struct()
export var RequestPaymentConfiguration = struct()
export var GetBucketRequestPayment = op()
export var PutBucketRequestPayment = op()

Although the bucketing doesn't create the absolute minimum import graph as individual files would, it balances the file count with importing a significantly and logically reduced subset of the schema tree.

The graph partitioning details are in the PR code.