bug(test): test_development_speedup swallows 3 of its 4 sub-test failures and still reports success #14931

test_development_speedup reports success when 3 of its 4 sub-tests fail

Found reviewing PR #14929. That PR converted the last of four swallowing handlers in this function into a raise; the other three still log and continue.

Duplicated verbatim in two files:

• autobot-infrastructure/shared/scripts/analysis/test_npu_code_search.py (~lines 144, 164, 184 still swallowing; 212 converted)
• autobot-npu-worker/npu_code_search_test.py (~lines 142, 162, 182 still swallowing; 210 converted)

The shape

try:
    duplicates = find_duplicate_code(...)      # Test 1
except Exception as e:
    logger.error(f"   Duplicate detection failed: {e}")   # swallowed, continues

Four such blocks: duplicate detection, pattern analysis, import analysis, comprehensive analysis. Only the fourth now raises. So if find_duplicate_code(), identify_code_patterns() or analyze_imports_and_dependencies() throws, the function still reaches its return True and the driver counts the test as passed.

That means the assertion #14929 added is narrower than it looks: it catches failure of the last sub-test only. Three quarters of what this test exercises can fail silently.

Why it was scoped out of #14929

#14929's remit was the driver-consumption defect — nine tests whose return True was converted to a bare return, breaking the driver's verdict. Un-swallowing the other three handlers is a behaviour change to what those tests assert, which is a different question and deserves its own review. Recorded here rather than folded in.

Note on duplication

These two files are near-identical copies. Whatever fix lands must land in both, or they drift — that is the [union-merge] concern that produced #14873 and #14886. Consider whether one should import the other rather than being maintained twice.

Acceptance

• A failure in any of the four sub-tests fails the test, not just the fourth
• The driver still distinguishes pass from fail (it sums truthiness over result = await test_func(), so the function must keep returning a truthy value on success — see bug(test): 262 tests return a value instead of asserting, so they cannot fail #14920)
• Both copies fixed, or the duplication resolved so there is one copy
• Mutation-proved: break each of the four sub-calls in turn and confirm the test goes red for each

Related: #14920 (returns instead of asserting), #14930 (the marker suite that cannot collect), #14873 / #14886 (duplicate modules that drift).