Treat entities with same type and value but dissimilar bodies as different entities?

[keshav47]

Describe the bug
Image this scenario:
We need to capture 10 digit mobile numbers uttered by the user from transcripts using ListEntityPlugin.

"alternatives": [
        [
            {
                "confidence": 0.9299549,
                "transcript": "my number is 9041270333"
            },
            {
                "confidence": 0.9299549,
                "transcript": "my claim is 1041270333"
            },
            {
                "confidence": 0.9299549,
                "transcript": "my claim is 1041270333"
            }
        ]
    ],

Passing the following pattern inside ListEntityPlugin:

entity_patterns:
  number_pattern:
    mobile:
    - \b\d{10}\b

Applying the current logic inside entity_consensus function, i.e

entity_type_value_group = py_.group_by(
            entities, lambda entity: (entity.type, entity.get_value())
        )

we capture the following (entity-type, value group):
{('number_pattern', 'mobile'): [KeywordEntity(body='9041270333', type='number_pattern', parsers=['ListEntityPlugin'], score=0, alternative_index=0, alternative_indices=None, latent=False, value='mobile', entity_type='number_pattern', _meta={})]

Due to this, we miss out on the other more recurring number i.e 1041270333.
In order to include this value, we can replace the above group-by logic with this:

entity_type_value_group = py_.group_by(
            entities, lambda entity: (entity.body, entity.get_value())
        )

this will capture the other entity like this:

{('9041270323', 'mobile'): [KeywordEntity(body='9041270333', type='number_pattern', parsers=['ListEntityPlugin'], score=0, alternative_index=0, alternative_indices=None, latent=False, value='mobile', entity_type='number_pattern', _meta={})], {('1041270333', 'mobile'): [KeywordEntity(body='1041270333', type='number_pattern', parsers=['ListEntityPlugin'], score=0, alternative_index=0, alternative_indices=None, latent=False, value='mobile', entity_type='number_pattern', _meta={})]

Disclaimer:
The above suggested group-by aggregation will not work with Datetime entities, can someone suggest a better alternative to handle this particular edge case.

[ltbringer]

I can support an expression in the candidate keys:
The current understanding is:

candidates = {
   "<entity_type>": {
       "<entity_value>": [
           ....
       ]
   }
}

patterns are assumed to have a value as their parent instead of the parsed value derived.

By supporting:

candidates = {
   "<entity_type>": {
       "__value__": [
           ....
       ]
   }
}

we can let the plugin understand that the parsed value should be used instead.