[hardware] 🐛 Fix vrgather out-of-range index hang in the mask unit

[emiliengnr]

Hello, here is a pull request for a bug I found.

Problem

A vrgather/vrgatherei16 with an index past the end of the source vector must
read 0 for that element (RVV 1.0 §16.4), and the mask unit correctly never requests
vd data for it. But the mask unit's issue gate still waited on
&masku_operand_vd_valid, so the gather never issued, the mask unit spun forever,
and the cva6 host stalled on the accelerator handshake until reset.

Minimal reproducer (NrLanes=2, VLEN=512, so VLMAX = VLEN/SEW = 64 at e8):

vsetivli x0, 8, e8, m1, ta, ma
li   t1, 200            # 200 >= VLMAX (64) -> out of range
vmv.v.x v8, t1          # all gather indices = 200 (OOB)
vid.v   v16             # source = [0,1,2,...]
vrgather.vv v24, v16, v8   # hangs

Fix

Waive the vd_valid wait in the issue gate when the index is out of range
(vrgat_idx_oor_q), for VRGATHER/VRGATHEREI16. The result path already
substitutes 0 for out-of-range elements, so no vd data is needed. Other ops are
unaffected.

Changelog

Fixed

• Fix hang when a vrgather/vrgatherei16 index is out of range

Checklist

• Automated tests pass
• Changelog updated
• Code style guideline is observed