Move structured import/export name information into the import/export string

[lukewagner]

This PR fleshes out the idea in #253 to move the structured information of imports/exports into the import/export name string, maintaining the same validation structure, but now inside a single quoted strong. The motivation for this change is to have a simple canonical string for all import/export names that contains all the structured information, without requiring host and toolchain implementations to define their own ad hoc string-serialization conventions. The rationale here is basically the same as the earlier design choice to put method/constructor/etc annotations inside the name strings, with the net effect of making the way externally-meaningful name information is represented in components somewhat more regular.

Some specific details:

• This PR is intended to be backwards-compatible at the binary format level, assuming that the 5 new "implementation import" cases added by #222 haven't been implemented or used yet. This binary compatibility is achieved by allowing a now-"junk" byte that was previously used to discriminate (interface ...) imports.
• The PR moves the existing kebab-name/label grammar to the "Imports and Exports" section so that the whole group can be defined in one block and also use a lexical grammar that's more explicit about whitespace and metacharacters. This ended up surfacing a spec-bug in which labels weren't being quoted.
• As there are now a fair number of "<___name>"s in the grammar, it seemed odd to have <name> refer to the kebab-name case. Moreover, <name> wasn't really a "kebab-name" anyway, since it also contains method/constructor/static annotations (with embedded square brackets and periods). Thus, this PR proposes to rename this production to <plainname>, attempting to distinguish plain names from all the other kinds of names that carry more semantic information, but happy to hear other thoughts on how to call this.
• Embedded URLs and registry names are deliminted by angle-brackets (not parens, as discussed in #253), mostly because (1) parens are allowed in URLs, (2) double-quotes require extra escaping in WAT, which seems like it'll be annoying, (3) other than double-quotes, angle-brackets are the RFC-recommended way to delimit URLs. The other implementation imports follow the same scheme as URLs just for symmetry.

Copying from this PR, an example showing the new import/export names is:

(component
  (import "custom-hook" (func (param string) (result string)))
  (import "wasi:http/handler" (instance
    (export "request" (type $request (sub resource)))
    (export "response" (type $response (sub resource)))
    (export "handle" (func (param (own $request)) (result (own $response))))
  ))
  (import "url=<https://mycdn.com/my-component.wasm>" (component ...))
  (import "relative-url=<./other-component.wasm>,integrity=<sha256-X9ArH3k...>" (component ...))
  (import "locked-dep=<my-registry:sqlite@1.2.3>,integrity=<sha256-H8BRh8j...>" (component ...))
  (import "unlocked-dep=<my-registry:imagemagick@{>=1.0.0}>" (instance ...))
  (import "integrity=<sha256-Y3BsI4l...>" (component ...))
  ... impl
  (export "wasi:http/handler" (instance $http_handler_impl))
  (export "get-JSON" (func $get_json_impl (result string)))
)

[alexcrichton]

To confirm, these two binary rules:

importdecl    ::= in:<importname> ed:<externdesc>         => (import in ed)
exportdecl    ::= en:<exportname> ed:<externdesc>         => (export en ed)

should change to refer to importname' and exportname', right?

[alexcrichton]

One other piece of feedback from https://github.com/bytecodealliance/wasm-tools/pull/1262, an implementation of this (and a few other PRs to the component model). Currently URLs are defined as "stuff that doesn't contain < or >", but integrity-metadata is defined with the w3c definition. It might make sense to do the same thing in both situations? Either require both are w3c-defined valid or both are something-nonempty-someone-else-parses-this.

I'll admit though I know nothing about integrity-metadata in w3c, if it's a simple format that's easy to parse then may as well "just" do that too.

[lukewagner]

@alexcrichton That's a good question and I was wondering about the same options for the same reason. Looking at the EBNF of integrity-metadata (here) and recursing through all the definitions, it does seem like integrity-metadata is much much simpler than a URL, but a good deal more complicated than, say, an interfacename, so it's in a sortof grey area where I could go either way.

But another reason in favor of having the Component Model fully parse integrity-metadata is that, unlike URLs, which are pretty universal (ignoring the many awful edge cases), there are, I believe, a couple of different hash text formats out there. E.g., OCI uses something different. Thus, it seems quite possible that different parts of the ecosystem might end up adopting slightly-different hash formats (accidentally or intentionally) if the common validation rules baked into the core-est tooling don't fix one.

[alexcrichton]

Ok makes sense to me! I'll dig in and implement that

[lukewagner]

Awesome, thanks to you and Alex for the feedback. I'll merge now since there don't seem to be any open issues and we can file follow-up issues if needed as usual.