Add task-based tracking to enable inference emissions estimation

[SaboniAmine]

First step to enable inference estimation estimation :

• Adding 2 methods to actual tracker : start_task & stop_task

Next steps may be to

• [x] Abstract the calls of those methods to a new decorator.
• [x] Add persistence methods to the task data type, and run the persistence at the tracker.stop() execution
• [ ] Define & handle production setups, for example when multiple workers run inferences on the same server instance. Would the PID be segregated ? If not, CPU, GPU & RAM measures may be duplicated between tasks.

[sashavor]

Some thoughts after testing the code:

• Can we report grams and not kilograms? Most inferences will only be a few fractions of a gram of CO2 in any case..
• It's a bit hard to debug code because of the logging the tracker is doing (took me awhile to find the error because it kept printing out the emissions every second). Can we make it crash if there is an error in the code itself, for example?

[benoit-cty]

Some thoughts after testing the code:

* Can we report grams and not kilograms? Most inferences will only be a few fractions of a gram of CO2 in any case..

* It's a bit hard to debug code because of the logging the tracker is doing (took me awhile to find the error because it kept printing out the emissions every second). Can we make it crash if there is an error in the code itself, for example?

Hello, you could change the log level in .codecarbon.config:

[codecarbon]
log_level = ERROR