## Summary

Restores `discv5` coverage to the devp2p Hive shard, which erigontech#21017
dropped from CI.

erigontech#21065 added `discv5` to the devp2p shard (`sim-limit: eth|discv5`),
closing erigontech#17354. erigontech#21017 ("matrix-test serial vs parallel exec across the
test workflows") then added a `serial`/`parallel` exec-mode axis and, in
its `ci: align test-hive devp2p sim-limit` commit, reverted both legs to
`sim-limit: eth` — citing a comment that erigontech#21065 had itself deleted. That
removed discv5 from CI entirely.

erigontech#21017's underlying point was valid, though: `discv5` exercises peer
discovery, not the EL execution path, so running it under *both* exec
modes would be redundant. So rather than re-adding it to both legs, this
keeps `discv5` in the **parallel** leg only and documents why the serial
leg omits it:

- **serial** devp2p leg: `sim-limit: eth`
- **parallel** devp2p leg: `sim-limit: eth|discv5`

`hive-versions.json` already carries erigontech#21065's pins (`hive_ref:
15750c9…`, `execution_apis_ref: 8d6b784…`), so `discv5` runs against the
same hive revision it originally passed on.

## Test plan
- Test Hive runs automatically on this PR via **CI Gate** — the changed
workflow file counts as a code change, so the `hive` job is not skipped
— exercising `devp2p` `discv5` in the parallel exec leg.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Co-authored-by: Giulio Rebuffo <111551070+Giulio2002@users.noreply.github.com>