Add NNCFProfiler implementation

[avolkov-intel]

Changes

Added implementation of profiler to collect and compare activations value of openVino model to investigate cause of accuracy degradation.

Reason for changes

Related tickets

162317

Tests

[ljaljushkin]

Thank you for the contribution, Anatoly!
Please consider adding visualization functionality and enhancing the scalability

[ljaljushkin]

First of all, it would be great to have visualization functionality as in the original scripts:
https://github.com/andreyanufr/compression_validation/blob/aanuf/activation_cmp/utils/cmp_activations/compare_model_outputs.py

Secondly, most likely there will be a need to add other comparisons and plots and keep it in one profiler class could be not extensible.
Could you think of registering comparators and visualizations inside profiler and delegate this responsibility to them?

For instance,

@comparator("mean_diff")
def mean_diff(a, b): ...
 
@visualizer("hist")
def hist_plot(...): ...

def compare():
    ...
    for metric in metrics:
        if metric not in COMPARATORS:
            raise ValueError(f"Unknown comparator '{metric}'. "
                             f"Available: {list(COMPARATORS)}")
        row[metric] = COMPARATORS[metric](a, b)
    
def plot(self, plot_type, *args, **kwargs):
        if plot_type not in VISUALIZERS:
            raise ValueError(f"Unknown plot type '{plot_type}'. "
                             f"Available: {list(VISUALIZERS)}")
        return VISUALIZERS[plot_type](*args, **kwargs)
        
cmp = profiler.compare(acts_fp, acts_int, metrics=["mean_diff", "relative_mean_diff"])

profiler.plot("hist", acts_fp, acts_int, layer="layer_7")
profiler.plot("mean_std", acts_fp)
profiler.plot("metric", cmp, metric="relative_diff")
# or alternatively have a different methods for plotting
profiler.plot_metric(cmp, metric="relative_mean_diff")
profiler.plot_hist(acts_fp, acts_int, layer="layer_10.attn")
profiler.plot_mean_std(acts_fp, act_type="in")