Remove flags from block_using_rules_sqls logic ( _find_new_matches_mode and _compare_two_records_mode etc.)

[RobinL]

Summary

This PR removes the following flags from the codebase:

• linker._analyse_blocking_mode
• linker.__two_dataset_link_only
• linker._find_new_matches_mode
• linker._compare_two_records_mode
• linker._train_u_using_random_sample_mode
• linker._self_link_mode
• linker._deterministic_link_mode

Instead, the required behaviour is achieved using function arguments primarily to block_using_rules_sqls

Details

At the moment a variety of boolean flags are set on the linker that modify the behaviour of sql generating functions.

It's difficult to understand the consequences of these flags because rather than the calling function 'asking' for specific behaviour by explicitly setting function arguments, the behaviour occurs implicitly via the flag.

For example, in linker.estimate_u_using_random_sampling, two modifications are made to blocking:

1. We need to take a sample of the input records
2. We need to eliminate blocking rules (because estimate_u_using_random_sampling performs a cartesian join)

But this behaviour is achieved by setting:
training_linker._train_u_using_random_sample_mode = True

and then in blocking.block_using_rules_sqls we have code that runs conditional logic like if linker._train_u_using_random_sample_mode e.g. here.

It would be much clearer if the estimate_u_using_random_sampling function explicitly took the samples, added the sqls to the pipeline, and then called blocking.block_using_rules_sqls passing in the name of the sampled tables and a null (full cartesian join) blocking rule.

Another example is how table names are interpolated into block_using_rules_sqls

e.g. here:

splink/splink/blocking.py

Line 251 in 4402316

from {linker._input_tablename_l} as l

Which has lines like:

splink/splink/linker.py

Lines 334 to 345 in 4402316

	def _input_tablename_l(self):
	if self._find_new_matches_mode:
	return "__splink__df_concat_with_tf"
	if self._self_link_mode:
	return "__splink__df_concat_with_tf"
	if self._compare_two_records_mode:
	return "__splink__compare_two_records_left_with_tf"
	if self._train_u_using_random_sample_mode:
	if self._two_dataset_link_only:

and is very confusing to work out what table name will be interpolated and why. Instead, within the linker.compare_two_records functions, it would be better to simply explicitly pass tablenames like __splink__compare_two_records_left_with_tf into block_using_rules_sqls

[ADBond]

Love this! Feels like a big QoL improvement for following the logic