This updates GoTool to subclass the newer BaseTool, and to use the

newer features it offers (dependencies, improved file extension

handling, new mixin for full-repository support).
It also fixes a number of issues that could arise with parsing the

output from both go test and go vet. The original implementation

requested JSON payloads from both, but go vet's JSON payloads are not

guaranteed. Parser/compiler errors would result in standard errors, not

JSON payloads, causing our parsing to break.
The new implementation just parses the standard error output,

simplifying the code considerably. That will catch both forms of results
The go test parsing has been improved as well. We no longer make

assumptions that all lines of code can be joined into a single JSON

array, instead parsing each line individually. We combine results

indicating failed test runs along with results showing test output into

a single representation, and build comments from that (helping

developers see why tests failed). We also provide general comments if

we fail to find any useful results but do find errors, showing the

output that occurred.

Unit tests pass on Python 2.7 and 3.x.