[alpha] Addition of simple pipeline abstraction

[NISH1001]

Major Changes

• evalem.pipelines module is added. Now a new type/component evalem.pipelines._base.Pipeline is introduced.
• evalem.pipelines.defaults.SimpleEvaluationPipeline is implemented which takes in a single model, input and list of evaluators to run the model through the pipeline. We can invoke .run(..) or the object itself is callable.

Minor Changes

• tests/ test suite is refactored to make use of conftest.py format configuration for pytest

Usage

from evalem.pipelines import SimpleEvaluationPipeline
from evalem.models import TextClassificationHFPipelineWrapper
from evalem.evaluators import TextClassificationEvaluator

# can switch to any implemented wrapper
model = TextClassificationHFPipelineWrapper()

# can switch to other evaluator implementation
evaluator = TextClassificationEvaluator()

# initialize
eval_pipe = SimpleEvaluationPipeline(model=model, evaluators=evaluator)

results = pipe(inputs, references)

# or
results = pipe.run(inputs, references)

Note: The pipeline is stateless. In future we can have another implementation that can take in the mappings of input datasets to corresponding models and produce some results for all the metrics.

[NISH1001]

Might have to pass kwargs to each evaluator just in case to avoid any bug?

[muthukumaranR]

you mean like a catch all kwargs that any evaluator might need, within the pipeline? if that's the case, it might be too overloaded imo. then again if we are not expecting too many evaluator specific kwargs, i think it's fine

[NISH1001]

Ah ya. For now, it's a bit tricky because some of the interfaces seem to lack kwargs. So, passing kwargs here explicitly might have runtime error. Gonna let it be for now.

[muthukumaranR]

not sure if testing reference and prediction lengths is a valid test but i might be wrong

[muthukumaranR]

I am not 100% sure about tests and their coverage - for now i think it looks good but leave it to @xhagrg @code-geek . rest(not much) is lgtm