Fix panic and wrong delete index in generated _SetVendor

[jamiesun]

Summary

The generated _<Vendor>_SetVendor helper (emitted into every vendor dictionary by radius-dict-gen) can panic with slice bounds out of range, and uses a wrong index when removing an emptied Vendor-Specific attribute. This fixes the generator template and regenerates the affected dictionaries.

This is the same crash reported in #121 (Juniper MX80 / ERX, also hit with 3GPP-IMEISV), and it was independently re-reported downstream in talkincode/toughradius#324, which additionally pinned down the i+i delete-index defect.

Defects (in dictionarygen/vendor.go, _<Vendor>_SetVendor)

1. Panic in the sub-attribute scan. The loop read the type/length at a fixed offset vsa[0], vsa[1] instead of the running cursor vsa[j], vsa[j+1], and unconditionally advanced j += int(vsaLen). After a matching sub-attribute is removed (vsa = append(vsa[:j], vsa[j+int(vsaLen):]...)) the slice shrinks, so advancing j runs past the end and len(vsa[j:]) panics. Reproduces by setting the same vendor attribute twice on one packet.

2. Wrong delete index. An emptied VSA was removed with p.Attributes[i+i:] instead of p.Attributes[i+1:] — a no-op at i==0, and dropping unrelated attributes (or panicking) at i>=2.

A third, coupled correctness issue: the len(vsa) > 0 write-back used copy(avp.Attribute[4:], vsa), which never shrinks avp.Attribute, leaving stale trailing bytes when a multi-sub-attribute VSA loses one entry.

Fix

• Read vsaTyp, vsaLen := vsa[j], vsa[j+1] at the cursor.
• Advance j only when the sub-attribute does not match (matching entries are spliced out in place).
• Rewrite the VSA at its correct new length: p.Attributes[i].Attribute = append(avp.Attribute[:4:4], vsa...).
• Delete an emptied VSA with i+1.
• Regenerated vendors/{aruba,microsoft,mikrotik,wispr} and rfc4679.

Tests

Added vendors/mikrotik/setvendor_test.go:

• TestSetVendorReplaceSameAttribute — the reported double-set panic.
• TestSetVendorDeletesEmptyVSAAtNonZeroIndex — the i+i out-of-range delete at index ≥ 2.
• TestSetVendorPreservesOtherSubAttributes — a VSA packing several sub-attributes is rewritten without stale bytes.

go build ./..., go vet, and the full go test ./... suite pass. Verified the new tests panic on the pre-fix code.

Relation to prior work

Thanks to @maddsua for reporting #121 and proposing #125 / #129. Those PRs change the loop bound to len(vsa)-j, which stops the panic but keeps the fixed-offset vsa[0], vsa[1] read and does not address the i+i delete index. This PR covers both, plus the write-back resize.

Closes #121