Refactor vdev_geom_attach and friends to reduce code duplication

[asomers]

Summary

Move checks for provider's sectorsize and mediasize into a single location
in vdev_geom_attach. Remove the zfs::vdev::taste class; it's ok to use the
regular vdev class for tasting. Consolidate guid checks into a single
location in vdev_attach_ok. Consolidate some error handling code from
vdev_geom_attach into vdev_geom_detach, closing a resource leak of geom
consumers in the process.

Test Plan

ZFS test suite

[asomers]

I declare a timeout on the CR. But by all means feel free to criticize it post-commit.

[avg]

In general I like this change very much as the code becomes safer and more readable.

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_geom.c

285

Could it happen that the acr was incremented by somebody else and we really shouldn't decrement the access bits?
I would be more comfortable if we passed an explicit flag to this function which would instruct whether we should "unaccess" rather than trying to guess that.

293

Just curious, in what case this condition could be false?

[asomers]

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_geom.c

285

I'll do it.

293

Uh, never? Looks like this line was copied by mistake.

[asomers]

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_geom.c

318

This assertion isn't always true in the case of spare vdevs, where two vdevs can share a consumer. I'll remove it. When I commit the change that turns cp->private into a linked list I'll restore a similar assertion.

[asomers]

Apply avg's comments. Remove a tautological assertion, remove an incorrect
assertion, and fix some theoretical null pointer dereferences in
vdev_geom_detach when vfs.zfs.debug=1.