WIP use more low-level API for control transfers

[tlyu]

Working proof of concept. This is kind of mixed up with trace logging code for now, but I'll clean it up in a bit.

Drawbacks include some stricter constraints on the callers of USBCore_::recvControl: the destination can't be of automatic duration, and the call will complete before the read actually occurs.

[tlyu]

On further reflection, maybe we should rename recvControl if we're going to choose this approach. The resulting API behavior is really different from the stream-based API of the AVR Arduino core. In particular, the destination buffer for the receive isn't filled until some unknown time after the recvControl call returns.

[obra]

For 3988c50, can you update it to talk about the why of the changes?

[tlyu]

For 3988c50, can you update it to talk about the why of the changes?

Done. I also took a closer look and decided that some of those changes were fine after all (avoiding changing a disabled endpoint). The change that really needed to be undone was only clearing the halt condition if the endpoint status was set to STALL.

[tlyu]

This is almost in a state suitable for use in an experimental snapshot. I've conditionalized the trace logging under -DUSBCORE_TRACE, so it won't slow things down when used with Chrysalis.