feat(util-endpoint): add endpoint ruleset cache

[kuhe]

aws/aws-sdk-js-v3#6423

This PR creates a cache for endpoint ruleset resolution.

As one would expect, the input and cache key are the endpoint params and the output and cache value is the endpoint.

Some services declare endpoint parameters in excess of what is actually evaluated in the ruleset, so codegen provides the explicit pared param list to the cache instance.

• codegen to connect EndpointCache to the service's endpointResolver.ts file contents.
• codegen to determine what the pared parameter list is to avoid cache pollution from irrelevant params

[kuhe]

The S3 resolver would now look like this below.
Notably absent is Key from the parameter list.

// smithy-typescript generated code
import { awsEndpointFunctions } from "@aws-sdk/util-endpoints";
import { EndpointV2, Logger } from "@smithy/types";
import { customEndpointFunctions, EndpointCache, EndpointParams, resolveEndpoint } from "@smithy/util-endpoints";

import { EndpointParameters } from "./EndpointParameters";
import { ruleSet } from "./ruleset";

const cache = new EndpointCache({
  size: 50,
  params: [
    "Accelerate",
    "Bucket",
    "DisableAccessPoints",
    "DisableMultiRegionAccessPoints",
    "DisableS3ExpressSessionAuth",
    "Endpoint",
    "ForcePathStyle",
    "Region",
    "UseArnRegion",
    "UseDualStack",
    "UseFIPS",
    "UseGlobalEndpoint",
    "UseObjectLambdaEndpoint",
    "UseS3ExpressControlEndpoint",
  ],
});

export const defaultEndpointResolver = (
  endpointParams: EndpointParameters,
  context: { logger?: Logger } = {}
): EndpointV2 => {
  return cache.get(endpointParams as EndpointParams, () =>
    resolveEndpoint(ruleSet, {
      endpointParams: endpointParams as EndpointParams,
      logger: context.logger,
    })
  );
};

customEndpointFunctions.aws = awsEndpointFunctions;

[kuhe]

In one benchmark, a batch of 10 parallel Amazon S3 HeadObject requests completed in 80ms, as opposed to 86ms without this cache.