Add string operator $replaceAll/$replaceOne tests

[eerxuan]

Add compatibility tests for the $replaceAll and $replaceOne operators.

Ref: #10

Tests cover:

• Core replacement behavior
• Encoding and special characters
• Type errors and invalid arguments
• Null/missing handling
• Size limits
• Invariants and type precedence
• Whitespace handling
• Expression arguments and document field usage

497 tests total (3 known failures: empty_find_multibyte tests fail due to pymongo UTF-8 decoding of invalid byte sequences).

[danielfrankcom]

Seems like we might need to add some kind of pytest mark to indicate these tests are expected to fail for MongoDB as previously discussed? If we override the merge conditions here we'll end up with all future PRs failing these checks which seems undesirable.

[danielfrankcom]

Won't the use of xfail mean that the compatibility tests expect all compatible databases to fail this test?

Since in this case it may be a database-specific issue, we may need a different approach here or we may cause this to fail for other compatible implementations even though the same problem doesn't exist.

[danielfrankcom]

Won't the use of xfail mean that the compatibility tests expect all compatible databases to fail this test?

Since in this case it may be a database-specific issue, we may need a different approach here or we may cause this to fail for other compatible implementations even though the same problem doesn't exist.

I'm not sure whether this would work for us in practice, but it looks like the xfail docs show a condition parameter. We may be able to isolate the xfail behavior only to MongoDB.

[danielfrankcom]

Seems like the condition needs to evaluate at setup time so accessing the engine name would be tricky.

Haven't tested this, but we might need to do something like this:

def pytest_configure(config):
    # We might be able to do this in the static `.ini` file instead, not sure.
    config.addinivalue_line(
        "markers",
        "engine_xfail(engine, reason): expected failure for a specific engine",
    )


def pytest_runtest_setup(item):
    for marker in item.iter_markers("engine_xfail"):
        if item.config.engine_name == marker.kwargs.get("engine"):
            pytest.xfail(marker.kwargs.get("reason", ""))

And then in the test file:

marks=(pytest.mark.engine_xfail(engine="mongodb", reason="MongoDB returns invalid UTF-8 for empty-find multibyte"),),

[danielfrankcom]

Most comments are around the introduction of xfail, which I guess is the downside of folding this change into an existing PR with test additions.