fix: browser respect child logger log level

[NicoVogel]

This PR addresses #960.

not yet completed, only a draft

Goal of this PR:

Child level determines which logs of the child are logged.

Implementation before this PR

• the prototype of logger contains the actual log functions which should be used
• the prototype log functions are assigned to the logger whenever the setLevel function is called
• for log functions which level is to low, noop is assigned instead

problem: only the main logger actually changes, even if the level of a child is changed

What this PR changes

• introduce a property which contains all the base logger functions (only the root logger has it) -> Symbol('pino.logFuncs')
  • the base log functions need to be callable independently by the child loggers and this property allows to access them all the time
• introduce a property which contains the hierarchy of the logger (pointing to the parent) -> Symbol('pino.hierarchy')
  • whenever a child logs, it needs all bindings as they are prepend to the log arguments.
    before this PR, it was handled via the prototype chain.
• move the responsibility of prepending the bindings into the set functions (for changing the log level)
  • as the set function is mainly responsible for defining the actual log functions that are usable, it felt like a logical step to also include the binding in it.
• add a bindings property to each child that stores the bindings of the child.
  • this is needed to walk up the hierarchy, get all bindings and then use it to prepend the log arguments.
  • there are restrictions to this property: it is only considered once, when calling the set function.
    so, reassigning it only takes affect if the level is changed and then it only is updated for this one logger.

Insights I want to share

There is a specific order in which a log is processed:

1. the outer log function is called with whatever should be logged
2. the bindings are prepended to the arguments
3. the LOG function is used to transform the arguments to the desired structure (can be found in createWrap)
4. the actual logging function is called with the transformed arguments

Before this PR, the prototype chain was used the ensure that the bindings are prepended one after another.
Now they are collected once and then prepended at once.

Another improvement that comes with this PR as well is, that now in case of noop the logger actually does nothing.
Before this PR, it always prepended the bindings and transformed the arguments no matter if the noop function was used or not.

there is a valid case as well and that is if a transmit exists, but this is properly handled

Finally, the base log functions are now only once defined in the setupBaseLogFunctions instead of in every set call.

[NicoVogel]

I am still not sure if the approach I take actually goes into the right direction.
My guess is that there should be a better way of solving this with the prototype chain, but I have a hard time getting into this mind set, because I never needed to work with prototypes before.

Anyone has a guess here?

[mcollina]

Yes, I like this approach.

[mcollina]

lgtm

[NicoVogel]

@mcollina I noticed that I used feat in the commit message instead of fix. Should I fix this or do you use squash commits anyways?

Furthermore, I am unsure how to proceed with the failing checks.

• the benchmark for current (master) failed, but I can't fix that in this PR
• 14 windows failed the installation of dependencies, but I did not touch the dependencies

As for windows 16 and 18, I will try to figure out why this fails

[NicoVogel]

@mcollina I noticed that no new release is yet available that contains the fix.
Is there still some work open?