build: remove unnecessary allocator-api2 dep.

[ErichDonGubler]

No description provided.

[zakarumych]

What is negative effect of the dependency?

[ErichDonGubler]

To directly answer your question: The presence of an unused dependency can be particularly burdensome in situations like Firefox's (i.e., in this patch stack where I'm revendoring WGPU), where trust for third-party dependencies are established by detailed review of each dependency's source. Each dependency in the tree is required to undergo significant scrutiny for the purposes of security, for each version. For what I hope are obvious reasons, when a dependency isn't actually used, our preference is to simply avoid the dependency altogether.

Taking a step back, your question surprises me. What user would want an unused-but-still-built dependency in their project? By default, I wouldn't expect a project using gpu-descriptor to ever be interested in the allocator-api2 crate, because its usage of hashbrown is not directly exposed in gpu-descriptor (which, unless I'm missing something, would be necessary for specifying collection-specific allocators). If a user does want this (i.e., via changing the global allocator in allocator-api2), they can still opt into it by selecting the appropriate features in their own direct dependency on hashbrown.

[ErichDonGubler]

CC @cwfitzgerald, who was a stakeholder for the hashbrown upgrade that I work with on a regular basis. He's also a stakeholder for this, since WGPU doesn't need this, either.

[zakarumych]

It's not like a user would want unused but still built dependency, but many users are not bothered enough to disable default ones.
My question was more like "how exactly it bothered you enough to go and submit a PR to disable it?"
And I thank you for thorough explanation.