[logger] `log.debug` and `.trace` should prefer the lazy implementation

[afharo]

There are many places where we do patterns like log.debug(`${JSON.stringify({...})}`) in the Kibana code.

When the log level is set higher than debug, this implies unnecessary CPU blocks only to be discarded later.

We should audit all use cases like this and change them to the lazy form: ``````log.debug(() => ${JSON.stringify({...})})```.

@dgieselaar provided the following RegExp to find all use cases:

\.(?:debug|trace)\(\s*(?!((?:\([^)]*\)|\w+)\s*=>)).*?JSON\.stringify\s*\(

Please, bear in mind that some loggers don't support this. Make sure to only update the ones using the Logger from @kbn/logger.

Ideally, an autofixable ESlint rule would prevent this from happening again.

[elasticmachine]

Pinging @elastic/kibana-core (Team:Core)

[jloleysens]

One alternative is to use if (logger.isLevelEnabled('debug')) {...}. It's maybe not as cool as lazy code 😉 , but it's intended to give callers a way to avoid even calling log for potentially expensive calls like JSON.stringify.

It would be nice to see if we can measure impact of these changes too.

[gsoldevila]

Here's a regexp that should work across multiple lines too:
\.(log|info|debug|error|warn|trace)\(([^)]|\n)*JSON\.stringify