Add per-engine expected result overrides for known engine-specific behaviors

[eerxuan]

Problem

Some test cases have known engine-specific behaviors where the expected result differs between engines (e.g., MongoDB vs DocumentDB). Currently we use pytest.mark.xfail to skip verification, but this:

1. Doesn't verify the test fails in the expected way
2. Won't alert us when the engine fixes the issue (xfail will silently start passing with strict=False, or break with strict=True)

Example

$replaceAll with empty find on multibyte characters — MongoDB returns invalid UTF-8 byte sequences instead of correct code-point-boundary insertion. The 3 affected tests (empty_find_multibyte_2byte/3byte/4byte) are currently marked xfail.

Proposed Solution

Add a per-engine expected result override mechanism to BaseTestCase, so tests can specify alternate expected values for specific engines. This would:

• Still run the test against all engines
• Verify the actual (buggy) behavior matches the known override
• Detect when an engine fixes the issue (override no longer matches → test signals for review)

This mechanism should align with the DocumentDB engine test infrastructure being built, so the same override pattern works for both MongoDB and DocumentDB differences.

References

• PR Add string operator $replaceAll/$replaceOne tests #79: $replaceAll/$replaceOne tests with xfail workaround