fix(e2e): deflake duplicate_proposal_slash by targeting offense type directly

[AztecBot]

Failing test on merge-train/fairies CI (run 26031248895, log http://ci.aztec-labs.com/1779103559110764):

FAILED (6b20f775668168e9): src/e2e_p2p/duplicate_proposal_slash.test.ts (309s)
  TimeoutError: Timeout awaiting duplicate proposal offense
  at duplicate_proposal_slash.test.ts:254:30

The merge-train/fairies branch only differs from next by three unrelated shell-script files (bootstrap.sh, spartan/scripts/install_deps.sh, yarn-project/aztec/scripts/add_crate.sh), so the failure is a flake.

The test previously did two sequential waits:

1. awaitOffenseDetected — wait up to 16 slots for any offense on honestNode1.
2. retryUntil — wait an additional 4 slots for a DUPLICATE_PROPOSAL offense across all nodes.

If a non-duplicate offense surfaces first on honestNode1, step (1) returns immediately and step (2) gets only 4 slots (96s) to find the duplicate-proposal entry. If the slasher's offenses-collector hasn't flushed it through within that window, the retry times out even though the duplicate eventually appears.

This collapses the two-stage wait into a single retryUntil that targets DUPLICATE_PROPOSAL directly with the full combined budget (20 slots), removing the race.

Investigation gist: https://gist.github.com/AztecBot/d6767ac7756f40cfc9e52157c8664ca1

ClaudeBox log: https://claudebox.work/s/a9d0c3aa5622d671?run=1