Require a box receiver for ssl/crypto's apply methods

[SeanTAllen]

Every apply on ssl/crypto's hash and one-shot primitives, and the HashFn interface, was declared fun tag. On a fieldless primitive tag and box compile to the same code, but tag forced a HashFn implementer to declare apply tag as well, because a box method is not a subtype of a tag one. Dropping tag widens what satisfies the interface and only breaks calling through a HashFn tag reference.

The interface has to lose tag before its implementers can, so all thirteen declarations change together.

Closes #94

[SeanTAllen]

Autonomous-mode decision log — PR-specific

No design decision here; the breaking change follows the issue's fix direction
directly. One correction the pre-PR review caught in my own work: the release
note first opened with "Every apply in ssl/crypto was declared fun tag,"
which is false — ConstantTimeCompare.apply and _EVPContext.apply were already
plain fun. Reworded before opening.