feat: xml serde reduction

[kuhe]

This PR implements a class called StringStore, which allocates shortened variables for any given string.

The basic idea is a transform of something like:

const body = '<?xml version="1.0" encoding="UTF-8"?>';
...
const body = '<?xml version="1.0" encoding="UTF-8"?>';
...
const body = '<?xml version="1.0" encoding="UTF-8"?>';
// in multiple functions

to

const body = _ve;
...
const body = _ve;
...
const body = _ve;
...
// usages only appear in functions so it's ok to define this at the bottom of the file.
const _v2 = '<?xml version="1.0" encoding="UTF-8"?>';

Real example from S3:

const headers = map({}, isSerializableHeaderValue, {
    "content-type": "application/xml",
    "x-amz-acl": input.ACL,
    "content-md5": input.ContentMD5,
    "x-amz-sdk-checksum-algorithm": input.ChecksumAlgorithm,
    "x-amz-grant-full-control": input.GrantFullControl,
    "x-amz-grant-read": input.GrantRead,
    "x-amz-grant-read-acp": input.GrantReadACP,
    "x-amz-grant-write": input.GrantWrite,
    "x-amz-grant-write-acp": input.GrantWriteACP,
    "x-amz-expected-bucket-owner": input.ExpectedBucketOwner,
});

const headers: any = map({}, isSerializableHeaderValue, {
    "content-type": "application/xml",
    [_xaa]: input[_ACL],
    [_cm]: input[_CMD],
    [_xasca]: input[_CA],
    [_xagfc]: input[_GFC],
    [_xagr]: input[_GR],
    [_xagra]: input[_GRACP],
    [_xagw]: input[_GW],
    [_xagwa]: input[_GWACP],
    [_xaebo]: input[_EBO],
});
// variables are defined at the end

To answer some potential questions:

• doesn't this make the code hard to read?

  • Yes. At least, you can hover over the variables to see what they are 😢. I believe users would prefer the reduced code size over readability of the protocol files.

• is the variable allocation deterministic?

  • Yes. Variables are named with a preference for the word initials of the given string. This makes the variables slightly longer than they need to be, but reduces variable thrashing with model updates.

• does this affect tree-shaking?

  • no. I tested with esbuild and only those variables being used are included in the bundle.

• what is the effect on protocol file sizes?

  • Note: this is not the size of the client, it's only the protocol file
  • S3 (plus other optimizations from feat: xml codegen reduction aws/aws-sdk-js-v3#5566) 412kb on disk to 328kb
  • EC2, 2.67mb to 2.5mb (there's room for further optimization)

• what about strings that only appear once?

  • within the codegen program's current framework, it is not known when allocating a variable how many times it will be used. It would require a second pass to inline variables with only 1 usage. This change increases char-count for strings that only appear once. However, in practice (spot-checking feat: xml codegen reduction aws/aws-sdk-js-v3#5566) most strings appear multiple times.

[kuhe]

requestBuilder is from #1107