Skip encoding mixing test in fuzzer when SkipValidation=true

[PranavSenthilnathan]

Mixing encodings will not throw when SkipValidation = true, so there's no need to validate that it does.

Closes #111588

[MihaZupan]

@MihuBot fuzz JsonWriter

[eiriktsarpalis]

Maybe I'm misunderstanding what this fix is doing, but the error in #111588 suggests to me that it's a misconfiguration issue, i.e. passing a non-expandable MemoryStream to a Utf8JsonWriter which is incorrect regardless of what inputs are being passed to the writer.

[PranavSenthilnathan]

The test is writing 2 segments of different encodings and asserting it throws. But in addition, it only provides a memory stream large enough only for the first segment, to validate that none of the second segment is written. In the fuzzer error, because validation is skipped, the second segment will be written anyway which exceeds the memory stream length.

[eiriktsarpalis]

Couldn't the fuzzer check that the total length is below the target threshold after writing has completed? The error message being surfaced right now seems a bit misleading to me.

[PranavSenthilnathan]

passing a non-expandable MemoryStream to a Utf8JsonWriter which is incorrect regardless of what inputs are being passed to the writer.

My understanding is that if the serialized output length is less than or equal to the non-expandable stream capacity then serialization should not fail. Is that not a guarantee we offer? This is a stricter condition than checks on BytesCommitted/Pending - for example, if the writer reports the expected BytesCommitted value but still writes more than it should then the test won't catch it.

[eiriktsarpalis]

I wouldn't disagree, but it just feels like an inappropriate way to enforce it. If we keep it like this I would prefer it if we inherited from MemoryStream and made sure a more appropriate error message was being surfaced.

[MihaZupan]

Should this one be merged?