Doc inaccuracy related to handler option WithCompression

[jhump]

The doc comment for WithCompression states:

Calling WithCompression with an empty name or nil constructors is a no-op.

However, tracing through the code, this is not true. The constructors are never checked for nil. Instead, any existing compressor previously registered with the same name is blindly overwritten with an invalid one. It is invalid because if either constructor is nil and ends up needing to be used, it will cause a nil dereference panic in the compressionPool here and here.

Perhaps this should instead behave like WithAcceptCompression? In that option, the constructors are checked. But if they are both nil, it effectively unregisters the named compression algorithm (instead of being a no-op).

[akshayjshah]

Yep, definitely a bug. Suggested fix makes sense - I think we just need to return a nil from newCompressionPool in this case and add a test.