Some source use client libraries that can emit errors that contain sensitive information - in particular, git-facing libraries that embed tokens into repository URLs. This PR introduces a way of redacting them - starting with GitLab (where we've seen this most recently), but in theory extensible to other sources as needed.
This implementation uses a custom zap core; this might also be possible with a custom zap encoder, but I didn't test it out.
(The deleted core.go file was entirely unused.)
* [THOG-643] Implement independent log level controls
There are two log level controls to mentally distinguish. Log levels
associated with a sink (e.g. stdout and streamed), and log levels
associated with a logger (e.g. a GitHub source).
The level is determined to be the minimum of the two. If a sink is at
level 0, then it will only output level 0 logs regardless of the
logger's level. This is best demonstrated by TestSinkWithName.
* Rename WithName to WithNamedLevel
* Check flush errors
* Replace IncreaseLevelCore with custom LevelCore
Adding a leveler that was less verbose would cause the initialization
fail, and therefore not be added to the core. This check is only at
the time of initialization.
An alternative approach to creating our own core is to set the child log
level equal to the parent, so initialization is guaranteed (with the
added benefit of intuitive behavior).
* Use controller if it exists, otherwise inherit parent's log level
* Cleanup some tests
