refactor: Remove optional feature attribute from stable feature gates

[vados-cosmonic]

With an eye towards making the discussion in component-model a reality, the commits in this PR update the wasm-tools to remove

Note that the change to wit-parser is a semver breaking change, as we're removing a piece of the Stability enum.

[alexcrichton]

Could this perhaps just be removed from enum Stability for now but left in enum Attribute? I'm wary to remove a feature cold turkey lest projects be broken, so having a smoother path to "this does nothing for awhile" before it's removed feels a bit better. Unfortunately there's not a great vector at this time to provide warnings about WIT files so we don't have the option of warning on deprecation of using WIT features at this time.

[pchickey]

For whatever its worth, I think for brand new features like this one, its OK to break this corner case. In a few months I'd be more careful, but I think its pretty unlikely anyone will encounter this.

[vados-cosmonic]

I can definitely see it both ways, and with regards to:

so having a smoother path to "this does nothing for awhile" before it's removed feels a bit better.

One thing I thought of might be to just make sure it the code never does anything useful (i.e. always puts None) and possibly outputs a warning.

So putting out the options at least:

• 1. Make feature always None (least breaking -- not a breaking change)
• 2. Remove from Stability, keep in Attribute (medium breaking)
• 3. Remove from both

Could go either way here, but if @alexcrichton you're comfortable with (2) I can go with that, or if you want to go with what @pchickey suggested (which is 3), I can do that as well.

[alexcrichton]

API-wise it's ok to have breaking changes whenever, that's what the semver-major-bumps are for in this crates on regular releases. Interface-wise-breaking though is more worrisome (e.g. changing WIT syntax). That's because it's expected that WIT with 215.* is compatible with 216.* and vice-versa (where possible). Changing WIT syntax by removing feature = ... breaks the both API and WIT compatibilty, but we can ignore the API parts.

That's basically a longer-winded way of saying, yes, I was adovcating for (2) and Pat for (3). That being said I don't disagree that this is new enough we can probably go with (3), so seems reasonable to just remove it outright (but that will require updating some tests it looks like)

[vados-cosmonic]

Yeah thanks for saying some more about it... I actually just looked at jco (where I thought there would definitely be changes), and changes aren't even necessary... Looking around a bit I can't find anyone really using the optional feature...