[rescue,test] enhance rescue error handling tests

[anthonychen1251]

This expands end-to-end tests for rescue error handling to increase test coverage:

• Adds a rescue_image_too_big test for Xmodem to check the handling of oversized firmware images during rescue, ensuring the device correctly cancels the operation and remains functional.
• Modifies the usb_dfu_out_chunk_too_big test to first perform a successful chunk download before attempting an oversized one, improving test robustness.

[pamaury]

Can you elaborate on this? As far as I know, there is no difference at the IP or SW level between the cw310 and the cw340 so it's not clear to me what is "unsupported" on the cw310?

[anthonychen1251]

This test aims to verify the USB IN transaction cancellation handling on the device side:

opentitan/sw/device/silicon_creator/lib/drivers/usb.c

Lines 333 to 335 in 382e702

	if (bitfield_bit32_read(reg, USBDEV_CONFIGIN_0_PEND_0_BIT)) {
	// Was there a cancelled transaction?
	error = kUsbTransferFlagsError;

In the rusb crate, I couldn't find a direct method in rusb::DeviceHandle to trigger a USB IN transaction cancellation event.

As a workaround, I implemented an approach where I issue a USB control request with a very short timeout (1ms) from the host side. In my development environment (cw340 board), this successfully triggers the desired corner case, and the device-side cancellation logic is exercised.

However, this test fails in the CI environment (cw310). It appears that for the cw310, this workaround of a minimal timeout request from the host doesn't lead to the device considering the transaction as cancelled. While I don't have deep insight into the hardware differences, I suspect this might be due to subtle timing differences and decided to only enable this test on cw340 platform.

[anthonychen1251]

I've observed that this test is flaky in the CI, even on the cw340 platform. I'm removing it from this PR for now and will investigate a more reliable way to cover this test scenario.