iflib: Fix panic when netmap VALE drops packets to iflib driver

[TeddyEngel]

When a netmap VALE switch drops packets destined for an iflib-based network driver (e.g., vmxnet3, igb, em), the kernel panics with a page fault in the driver's isc_txd_encap() function.

The crash occurs because:

1. nm_vale_flush() drops oversized packets by setting slot->len = 0
2. iflib_netmap_txsync() processes the zero-length slot but seg_idx stays 0 since len == 0
3. isc_txd_encap() is called with pi->ipi_nsegs = 0
4. The driver loops zero times but unconditionally sets the EOP flag on an uninitialized descriptor pointer

To fix, we skip zero-length packets (seg_idx == 0) at end-of-packet. Advance nm_i to consume the netmap slot but not nic_i, avoiding gaps in the NIC descriptor ring.

The fix is in iflib rather than individual drivers because all iflib drivers assume nsegs > 0, and we want to protect all drivers with a single fix.

Tested on FreeBSD 16.0-CURRENT using the reproducer script from the bug report.

PR: 294726

[TeddyEngel]

CI failures seem unrelated to this PR - they're in lib/libc/gen/getcap.c (conflicting types for cgetent/cgetfirst/cgetnext). Same failures also appeared on PR #2263. Style Checker and commit checks pass.