net80211: add a TX mbufq to ieee80211_node to be managed by drivers

[bz]

Summary

For recent drivers we run into issues on TX in that we can
malloc/sleep. For the moment rather than replicating some logic
into drivers or compat framework start migrating a per-node mbufq
into net80211 and let the driver manage it. This allows the driver
to defer the TX for now avoiding locking issues.
Eventually the net80211/driver TX interface should improve and
most of the needed should be done in net80211.
I am just not there yet as to know what exactly we will end up with
so taking it one step at a time.

Also add a drv_data pointer where the driver can store state,
e.g., to backtrack its own internal view on the node.

Sponsored by: The FreeBSD Foundation
MFC after: 2 weeks
X-MFC: wait longer, I just want the email

[bz]

@adrian from my understanding this would also help "athp" in one of its recent incarnations to get some FreeBSD specific code out of the driver.

I am not entirely happy with this but keeping it stuck in a development branch will not help the discussion either so putting it out here.

[adrian]

Yeah, I was thinking the same thing. It gets a bit unfun when we start looking at how we drain this versus direct dispatch, but let's tackle that later.

(ath and athp have the same thing.. ugh)

[sam]

I assume there's more coming that uses this interface? Without seeing how this is gonna be used I think you should hold onto this.

[bz]

I'll add the ni_drv_data pointer but apart from this I am abandoning this.
Doesn't seem right to have it there anymore.
A lot more to figure out properly first.