Stack Overflow during TextFormat generation

[thomasvl]

oss-fuzz found a test case that is properly parsed, but fails to re-serialized it back out to TextFormat, but only in the debug version of the fuzz test. The input is:

79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{49{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{49{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}49{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}49{79{79{79{79{79{79{79{79{79{79{79{79{79{79{79{}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}}#49{79{79{7}{7

Which basically boils down to just deep nesting of the recursive message, close to the recursion limit.

This sorta appears to be like #1034.

Since the TextFormat output is used for the debugDescription not being able to dump a deep message that were were able to parse will likely be a little confusing to developers that run into this.

[thomasvl]

I tried a manual edit to the generated fuzz_testing.pb.swift version of traverse to wrap each if let ... block with a try { if let ... }() closure, like we did for the other issue. That does seem to reduce the stack usage in debug build so the a rerun of the data on FuzzTextFormat passes.

[thomasvl]

Fixed via #1183