fix(tests): fail loudly if cost model already updated

[OlufemiAdeOlusile]

Summary

The upgrade test is meant to prove three states: old protocol
version, new protocol version with cost model still short, new
protocol version with cost model fully updated. The middle state can
currently be skipped without anyone noticing, because
test_update_cost_models always proposed the given cost model
values, even when the live cost model was already at or above the
proposed length. That silently loses the "not yet updated" scenario
while still reporting a pass.

Fix

Added is_cost_model_update_needed() in conway_common.py, a
standalone check separate from update_cost_model(). It compares
live cost model length against a proposal file for a caller-given
set of model names, and returns whether an update is still needed.
update_cost_model() itself stays update-only, no gating logic
inside it.

test_update_cost_models (step2 of the upgrade test) now asserts an
update is needed before proposing one, and fails loudly with a clear
message if not, instead of silently re-proposing values that don't
change anything.

test_update_in_pv9 and the PlutusV3 builtins test already had
their own pre-checks before calling update_cost_model(), so they
don't need to change.

Fixture fix

cost_models_list_185_297_v2_v3.json targeted PlutusV3=297, stale
since cardonnay PR #148 (merged) bumped local_slow's Conway genesis
straight to the current mainnet value (350). Updated the fixture to
match.

Verification

• ./ai_run.sh make lint clean
• Ran the real end to end upgrade test (all 3 steps) with these
  changes. All green, zero failures. test_update_cost_models
  correctly found the cost model still genuinely short at that
  point, ran the real governance update, and passed.