The logs on the node, no longer display the 'original' query, they only display the final 'idempotent' query. (#366)

Histograms fix bins

MIP-674 Anova

- Two-way ANOVA implementation [MIP-674](https://team-1617704806227.atlassian.net/browse/MIP-674).
- Two-way ANOVA validation. This is currently incomplete as there is a bug in the computation of the degrees of freedom. [MIP-675](https://team-1617704806227.atlassian.net/browse/MIP-674).
- New `FormulaTransformer` class in `algorithms/preprocessing.py`.

Cross validation algorithms are tested probabilistically by means of a
t-test. This means that all tests are not supposed to pass, since they
are probabilistic. This is a problem in our case since we need all tests
to pass. Unfortunately the only way to resolve the issue is to cheat
and replace the test with another one, hoping the new one passes.

https://team-1617704806227.atlassian.net/browse/MIP-675

Also some results are added to LinearRegression class because they are
used in ANOVA.

https://team-1617704806227.atlassian.net/browse/MIP-674

A FormulaTransformer class is implemented. This class takes an R style
formula and a matrix X and transforms the matrix into a new one,
possibly with a different schema, with columns defined by the terms of
the formula.

The implementation relies on the patsy library.

When a new function is stored in the UDF registry, a unique name is
generated based on the function and module names. This resulted in name
collisions, since one module could host different classes with the same
method names.

This is now fixed by using the qualified method name instead of the base
name.

* all algorithm modules must implement the Algorithm abstract class
* create_primary_data_views method removed from AlgorithmExecutorInterface, db views are created before starting executing the algorithm flow code

Fixed quart backwards incompatibility of 'serving_handler'.
When hypercorn was added the logs stopped working properly. The 'accesslog' parameter was needed.
Added configuration for default hypercorn loggers.
Centralized all the loggers into one file so it can be loaded from multiple places, with/without hypercorn.
Configured the hypercorn 'access_log_format' to see when the controller api is used.

Only the patch version will be increased by poetry update.

In this mipdb version the pin of pymonetdb package to 1.4 was removed and it was updated to 1.6.3

Pinned numpy, scipy, pandas, scikit-learn and statsmodels versions to latest, the same as in the monetdb Dockerfile. Added comments as well to avoid difference.
Updated black and isort packages, in pyproject.toml as well as in github actions.
dnspython was pinned to 2.2.1 due to a bug in eventlet https://github.com/eventlet/eventlet/issues/781 .

The methods were combined in order to create the table with only one query instead of several.
Since the 'create merge table' is now only one, it can be converted to idempotent query.
Changed the 'create_fault_tolerant_query' to 'create_idempotent_query' since it showcases the meaning better.

All the tables that are now created or deleted, if the connection fails and the query succeeds, can be retried without a 'Table already exists' problem.
In the 'monetdb_interface' moved some global variables inside the methods to mock properly.
Removed soft memory limit from docker containers.
Added unit and integration tests testing the 'IF NOT EXISTS' or 'IF EXISTS' clause in the table queries.

The drop query was added as part of the udf execution queries to be able to retry the series of queries even if it previously failed at some point.
This is no longer the case since we retry each query, if it failed.

* Added very_slow mark on the tests that stop/remove/restart/kill containers.

* Added smpc_cluster mark on the test that setup the smpc cluster

* Disabled the smpc cluster tests cause of its instability

* New mipdb version.
Mounted the csvs on the monetdb of Kubernetes.

Copied the tests and the expected of the algorithms leaving only the first 5.

algorithms: reduce memory footprint of local UDFs