Source piping

[tea-dragon]

It almost certainly needs at least a little bit more work, but if someone wants to throw out some feed back, that could be nice.

[abramsm]

• please add class comments to StreamSourceByPaths
• LinkedHashMultimap is not thread safe but we do not appear to have any protection against concurrent calls. Is there something higher up making that guarantee?
• tests for MeshyStreamFileComparator would be good
• in StreamSourceMeshPipe the use of private and package local seem somewhat arbitrary.
• class documentation for StreamSourceMeshSearch

- can we break 'findMeshFiles' into a few smaller methods for better [read,test]ability?

[tea-dragon]

• agreed about comments. Need them for most classes.
• the class needs to be marked as 'not thread safe'. There is something higher up making that guarantee (the class is basically just an encapsulation of logic that used to live in StreamSourceMeshy. No concurrency changes have been made).
• MeshyStreamFileComparator is actually the only class that is tested, but I didn't change the name of the test from TestStreamSourceMeshy. Also, I think the test might only check for exceptions. Tests for the other classes would be nice.
• Only two package local scopes are for two unused methods that represent logic that used to exist but was broken or redundant. They needed package local when I initially extracted other classes that were either inner classes or part of StreamSourceMeshy. They are package local because they used to be essentially private, but since the mesh package was added and consists mostly of things that used to live inside StreamSourceMeshPipe, this more or less preserves their previous scope. They could be removed, but I left them in for now so they wouldn't be forgotten entirely. Prior to merging i'll clean them up one way or another.
• agreed as previously mentioned
• agreed on breaking up findMeshFiles, but being able to create a distinct StreamSourceMeshSearch object without an enclosing StreamSourceMeshy, and particularly with a non-codec constructor is already miles more test/read-able than it used to be.

[abramsm]

this looks good to me. May have drifted a bit from master. Has it been tested in test cluster?