[arkadiuszwojcik]

Both bugs were introduced in 91773f5 ("Update code for new @bitcast semantics"). On a Pico 2 W running net-pong, the first made every event get dropped, so the join timed out. The second made every data frame get dropped, so DHCP never completed.

ioctl.zig: skip BDC padding in Response
Since the port, Response.data() returns reader.buffer[seek..end], but bdc() consumed only the BDC header and not the offset * 4 padding after it. For events with bdc.offset != 0 the payload started too early, event() returned null, and all events were dropped. bdc() now also discards the padding. data() is unchanged, and a bogus offset leaves data() empty. Added a regression test with a captured event packet.

wifi.zig: send_zc bus write start
send_zc passed as_words(writer.buffered(), …), which starts at buffer + 4, so bus.write's command word overwrote the SDPCM len/not_len and the chip dropped every data frame. It now passes as_words(buffer, words_len) again, as before the port.

Verified on a Pico 2 W with net-pong: the join succeeds, DHCP leases an address, and the board answers ping.