test(ci): drop e2e_kernelless_simulation from backwards-compat e2e

[AztecBot]

Summary

Forward port of #22999 (which targets backport-to-v4-next-staging) onto merge-train/fairies so the v4 and v5 versions of compat_test_cmds stay aligned.

Excludes src/e2e_kernelless_simulation.test.ts from the compat_test_cmds set in yarn-project/end-to-end/bootstrap.sh, so it no longer runs under the backwards-compatibility (ci-compat-e2e) job.

Why

e2e_kernelless_simulation asserts on the exact number of nullifiers emitted by the test contracts. PR #22939 changed the contract artifacts and now emits a different nullifier count, which causes the test to fail when the new client runs against older contract artifacts in compat mode (failing v4 nightly: https://github.com/AztecProtocol/aztec-packages/actions/runs/25419116073/job/74558302821#step:4:38). The mismatch is not a real backwards-compat regression — the test is asserting on internal accounting that legitimately changes between versions.

Per @benesjan and @mverzilli, the right call is to drop this test from the compat matrix rather than maintain compat-aware nullifier-count assertions; reducing the compat surface to tests that actually exercise the contract API is the longer-term direction. Forward-porting here keeps the v5 compat matrix in lockstep with v4.

Implementation

Identical to #22999: add kernelless_simulation to the bash extglob exclusion list for the top-level e2e_*.test.ts pattern.

-    src/e2e_!(block_building|prover_*).test.ts
+    src/e2e_!(block_building|prover_*|kernelless_simulation).test.ts

Comment block above the function updated with the same reasoning. Verified with shopt -s extglob that the new pattern excludes only src/e2e_kernelless_simulation.test.ts and still includes every other top-level src/e2e_*.test.ts.

ClaudeBox log: https://claudebox.work/s/ed6536a05f7d2b77?run=4