fix: return ErrMessageSizeTooLarge when message is too large

[ae-govau]

For most of this library's existence it has returned ErrMessageSizeTooLarge when the message exceeded the configured size.

For a short period in 2019 this error was renamed (#1218) but shortly revered back (#1262). Later in 2023 this error was changed to a ConfigurationError (#2628) to fix #2137, however this has caused issues with clients who rely on the previous error code being distinct from other ConfigurationError conditions (#2655).

This commit reverts to previous behaviour, and adds a test to pickup if this changes again in the future.

[ae-govau]

@puellanivis - this might be a better approach? I know you like the errors.Is() but I added the equality check for the err in the tests as I suspect that's what most code-bases currently do.

@shermanCRL , @dnwe and @hindessm who reported #2137 and made #2628 which this effectively reverts who may be interested in this discussion and have comments.

An altnernate PR which doesn't change the current types returns is proposed in #2848. Personally I prefer this approach in this PR (ie restore previous behaviour), but that unresolves #2137 which may or may not be OK.

[puellanivis]

I mean, one of the first tests errors.Is does is check to make sure the errors are comparable and then err == target. So, testing with equality here isn’t so bad (and is a stronger assertion for the package API tests, that we’re returning directly the ErrMessageSizeTooLarge and not any wrapped version.

[puellanivis]

To be more clear, I’m more behind this revert, rather than adding a new function to test an error string “safely”.

[ae-govau]

Pushed additional commit to fix the lint. Force-push because I forgot the DCO the first time.

[ae-govau]

It's not clear to me why that one test failed in one env. Any chance that's a flaky test?