Timestamps are not accurate if bus is suspended and resumed during the capture

[desowin]

Copied from openvizsla#55

Packet timestamps are generated based on 24 bit 60 MHz counter value. For example OutputPcap generates nanosecond resolution timestamp based on the counter value:

ov_ftdi/software/host/ovctl.py

Lines 221 to 231 in f70968b

	# Increment timestamp based on the 60 MHz 24-bit counter value.
	# Convert remaining clocks to nanoseconds: 1 clk = 1 / 60 MHz = 16.(6) ns
	if ts < self.last_ts:
	self.ts_offset += (1 << 24)
	self.last_ts = ts
	clks = self.ts_offset + ts
	if clks >= 60e6:
	self.utc_ts += 1
	self.ts_offset -= 60e6
	clks -= 60e6
	nanosec = int((clks * 17) - (clks // 3))

The approach of detecting counter overflows by comparing timestamp against previous timestamp works fine when there is actual bus traffic (even when there is no transfer in progress there is a steady number of SOF packets sent on Full- and High- speed links). However, the timestamps become no longer relevant when the bus is suspended and therefore there is no traffic whatsoever on the bus. Host has no way to tell how many overflows there were during the suspend.

A naive solution is to send new magic byte on every counter overflow. Then the host software could use it to keep the timestamp accurate while the bus is suspended.