[IRGen] NFC: Add operator<< for MetadataResponse

[davezarzycki]

Newer versions of clang are more reliable about emitting "used" functions, but that means that newer versions of clang can break the build if the dependencies of a "used" function aren't satisfied.

[davezarzycki]

@swift-ci please smoke test and merge

[rjmccall]

Was this used somewhere? I don't entirely understand the point of adding a debugging dump that doesn't seem to print out the state very faithfully.

[davezarzycki]

If you look at #27950, you'll see that newer versions of clang faithfully emit an "attribute used" dump() function whereas older clangs might not. This means that the dependencies of the attribute used function (like operator<<) need to be there.

I CCed you on this bug because I thought you might care about the QoI of this output operator.

[rjmccall]

No, I understood that there was a use of this that you wanted to satisfy. I'm asking what that use was from.

[davezarzycki]

Not anything more specific than including a header as far as I can tell. The clang error "backtrace" is in #27950 is what I observed and it is less than perfect/helpful.