Avoid overwriting HCI buffer until sent

[bjoernQ]

When implementing an BLE HID device it turned out that for BTDM based drivers we shouldn't touch the packet passed to API_vhci_host_send_packet until it's sent

API_vhci_host_check_send_available doesn't seem to be suitable for that - for an easy fix this uses the notify_host_send_available callback to toggle a flag which seems to fix the issue

NPL based drivers seem to copy the packet in their send function and are not affected

If this ever turns out to become a performance problem, we can implement something more advanced but for now it seems to be good enough

[MabezDev]

LGTM, thanks!

[bugadani]

I guess it's too late but I'd have preferred waiting right before PACKET_SENT.store(false, Ordering::Relaxed);, so that the code doesn't unnecessarily burn CPU time.

[bjoernQ]

I guess it's too late but I'd have preferred waiting right before PACKET_SENT.store(false, Ordering::Relaxed);, so that the code doesn't unnecessarily burn CPU time.

I thought about it but then it's possible that code will overwrite the in-flight packet before it's completely processed

[bugadani]

ah okay, makes sense I guess