Rename *metrics collectors* to *publisher* when inference scheduler is the source

[elevran]

#414 introduced mechanism to enable additional metrics from inference-scheduler.
The PR used the term metrics collector, possibly to align with the Prometheus naming conventions.
However, from the inference scheduler's point of view it is exporting/exposing/publishing additional metrics and not collecting them (e.g., metrics collected from inference servers to guide scheduling).
Consider changing the concept docs and code to reflect this.

[nirrozenbaum]

cc: @JeffLuoo

[JeffLuoo]

Collectors are eventually registered under to a central collector registry according to prometheus guide: https://prometheus.io/docs/instrumenting/writing_clientlibs/#overall-structure

I think using the term "collector" should be fine here but open to rename most mentioning of "collector" to "publisher" if you think it fits better in the stack.

[elevran]

Agree that the term collector is fully aligned with and warrented by the Prometheus model and documentation.
In the context of inference schedule work (and not having developed with Prometheus before), I found the directionality implied by collector counter intuitive.
My preference would be to use "publishing" and derivatives over "collecting" for this context.
However, I don't feel stronly about it, so would let others weigh in.

[shmuelk]

After thinking about some more and looking at the Prometheus APIs a bit. Collector seems to be the correct word.

[elevran]

Thanks @shmuelk
Unless there are other objections, will close this as invalid/won't fix and use Collectors as in #414.