[RFC] New way of handling frame IRQ

[marcio-ao]

Currently, the frame IRQ fires continuously every 1ms. This causes problems with timing critical code, especially code which also has a 1kHz interrupt (I'm talking about Marlin).

This PR modifies the handling of the frame IRQ so it is enabled only when needed:

• The sof_countdown variable has been refactored into a SOF_Timer class
• The SOF_Timer now has a "woken" and a "sleeping" state.
• A device insertion or removal causes the CONDETIRQ to fire
  • The IRQ calls "wake()" on the SOF_Timer
  • This starts the 1ms timer, by setting the FRAMEIE bit in HEIN
• State transitions continue at 1ms intervals until a USB_RUNNING, USB_ERROR or USB_IDLE state is reached
  • The state machines calls "sleep" on the SOF_Timer
  • The SOF_Timer clears the FRAMEIE bit in HEIN, silencing further frame interrupts and freeing up the CPU to do other things
• While the timer is "asleep", no polling happens, but the CONDETIRQ will fire in the case of a device removal or insertion.
  • This will "awake" the SOF_Timer for processing of the event.

Don't merge this code in quite yet. I believe it can be cleaned up a bit more. I just wanted you to look over it and maybe test it on your devices.

[xxxajk]

Please refactor, then I can review it. See also private email on concerns and validation testing.

[xxxajk]

One other note: there are other host controllers that would have to be modified, not just the MAX chip. EHCI and Kinetis, plus anything in the future. Might be best to have a variable or a define to enable this feature as well.