test(bb): avoid vacuous iterations in BatchMultiScalarMul (merge-train flake refresh)

[AztecBot]

Why

The merge-train PR #23137 (head d016756552) failed its merge-queue CI run on a TXE/AVM flake unrelated to anything on the train — noir-contracts orderbook_contract test::test::create_order_zero_ask_amount got Timeout waiting for TypeScript callback for contract class from the AVM NAPI callback after 60s. The PR that was just merged into the train (#23226) only touches one barretenberg MSM test file, so it cannot have caused the failure. Per .claude/skills/merge-trains/SKILL.md option 1, landing a small PR onto merge-train/barretenberg refreshes the HEAD SHA so auto-merge can re-engage on #23137 (it won't re-enable after a failure conclusion).

Full analysis (failed run, log keys, why we can't reproduce ./bootstrap.sh ci locally): https://gist.github.com/AztecBot/24eb9bfac0fb6e29bdfcf031a7f50a63

What

Tiny test-quality follow-up to #23226 in the same file. test_batch_multi_scalar_mul was computing num_msms = rand_u8 % kMaxBatchMSMs and num_pts = rand_u16 % kMaxBatchPointsPerMSM, both of which can be 0 — making that iteration (or the whole test invocation) vacuous. Switched to 1 + (rand % (kMax - 1)) so every run exercises at least one MSM with at least one point. Max bound is unchanged (still kMax - 1), so the existing static_assert(kMaxBatchMSMs * kMaxBatchPointsPerMSM < num_points, ...) and runtime ASSERT_LT continue to hold.

How to land

This needs the ci-skip label and a force-merge via "Merge without waiting for requirements to be met" — claudebox can't apply labels itself. The point of this PR is to refresh the merge-train SHA, not to validate the test-quality tweak via CI.

Verification

cd barretenberg/cpp && rm -rf build && cmake --preset default -DBB_LITE=ON
cmake --build build --target ecc_tests
./build/bin/ecc_tests --gtest_filter='*ScalarMultiplicationTest*BatchMultiScalarMul*'

6/6 pass on BN254 and Grumpkin (BatchMultiScalarMul, BatchMultiScalarMulSparse, ScalarsUnchangedAfterBatchMultiScalarMul).

ClaudeBox log: https://claudebox.work/s/c5fe8d2727b143f9?run=1