usb: recover endpoint 0 after a failed request; report-protocol keyboards

From the first boot on real hardware: a Razer mouse's control transfer
failed with a babble, and an ITE controller's late endpoint-0 completion
was reported as an "unexpected event" during a command.

A failed control transfer left endpoint 0 halted (or, on a timeout, still
running with TRBs queued), so every request after it to that device failed
too. It is now reset or stopped and its dequeue pointer moved past what was
left on the ring. A STALL -- a device declining a request it does not
implement -- is logged only on a verbose boot.

HID interfaces without the boot subclass were turned away unless they had a
pointer. The report descriptor's keyboard page is now read too -- modifier
bits, a key array, an NKRO bitmap -- and each report rewritten into the boot
layout for the existing path. Checked with QEMU's usb-kbd forced into report
protocol: "echo Hi", Shift included, typed and ran.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01S1giNvrHTrPMPw7V6x1kQ4

kernel/drivers/bus/xhci/xhci.h

@@ -143,6 +143,8 @@ struct trb {
 #define TRB_CONFIG_EP       12
 #define TRB_EVAL_CTX        13
 #define TRB_RESET_EP        14
+#define TRB_STOP_EP         15
+#define TRB_SET_TR_DEQ      16
 #define TRB_NOOP_CMD        23
 // Event TRBs
 #define TRB_XFER_EVENT      32
@@ -151,6 +153,8 @@ struct trb {
 
 #define TRB_CC(status)      (((status) >> 24) & 0xFF)
 #define CC_SUCCESS          1
+#define CC_BABBLE           3
+#define CC_STALL            6
 #define CC_SHORT_PACKET     13
 
 // Setup Stage transfer type, and Data Stage direction.