…s. Fix some warnings.

May be needed in the future for NITTA profiling purpose.

Regex-posix now have their own dependency for regex-posix-clib on
Windows.

Related to #235

Due to some platform-specific peculiarities (I guess) exiting run_nitta
context manager is killing the whole process group including the
parent process of the Python interpreter itself.

Python tests were killing JetBrains IDE components in Docker
development environment because of this 😅

Updates haskell code to reformatted with recent fourmolu

Failed to do so via StateT monad at the moment due to rising problems
with type constraints ("could not deduce...") in Main.hs or when
synthesis methods were used outside of NITTA REST API.

…ototype

See #240 (comment)

- omitting uid/gid
- ghcup
- unminimizing + man
- gpg passphrases
- fix PATH and .bashrc issue
- move aliases to .bash_aliases
- fourmolu 0.10.1.0

Refactor everything to Pydantic for consistency

- Python development was switched to VS Code since JetBrains Gateway
is still too buggy and painful to use for remote development over SSH.
- All Python code was reformatted with `black`.
- All Python code was linted and made `mypy`-compatible.
- All imports in Python code were regrouped and sorted with `isort`.
- All unused imports were removed with `autoflake`.

Besides:
- Basics for tree sampling were implemented. Yes, this should've been
a separate commit but things got messed up :c

It's important in async scenaries when multiple coroutines are working on
the same node. One might incorrectly consider it already loaded.

- Label calculation
- DataFrame assembly
- Saving to csv

- Implemented label aggregation and sampling results deduplication
- NITTA was failing to start sometimes due to port conflicts - fixed
- Implemented sampling tree coverage estimation (phew...)

Doesn't work for other examples and I've run out of time for that :(

Aggressive drops jump straight to top-n node from the heap (instead
of top-1) when synthesis is beating around the bush for too long. This
was found to be helpful for huge synthesis cases.

ML for synthesis prototype v2 -> v1 branch

It was rare but sometimes it happened

This is a fix for a "port is already in use" occasional problem with
dynamic startup of NITTA API server in Python scripts. Before we've
chosen a random port on Python side and passed it to NITTA via --port.
Sometimes it was already occupied by another process. That caused
random crashes.

This could be fixed more easily by startup retries with port reselection
if only we didn't need to support API server startup right after
synthesis. Synthesis can take a long time, hence greater probability for
chosen ports to become occupied and greater time cost of the retry.

The only robust solution was to implement a NITTA-side free random port
selection and its passing over to Python via inter-process communication.
This commit implements it via Python-side NITTA stdout parsing as this is
the simplest way to achieve the goal.

The scripts now uses the updated behavior of NITTA itself (post-synthesis
API server startup) and dynamic NITTA startup from Python (NITTA-side
random port selection).

Metrics gathering that was earlier implemented via NITTA stdout parsing
is now fully `/treeInfo`-based.

- sum/min/max/mean when len in (0, 1)
- mistakenly taking columns only from results[0],
although it can be a timeout (hence not a full dict).

Successful leafs count wasn't checked, so non-timeout synthesis failure
wasn't detected. NITTA API server was started anyway in those cases, so
it was considered a success and led to stats evaluation errors and
incorrect results.

Evaluation showed that "STEP LIMIT REACHED" was sometimes happening even
before a 60 seconds synthesis timeout was reached.

They were uploaded just to get a permalink and are not actually needed
in the repo.

This change replaces the old full-tree approach of gathering model
training data with the recently developed tree sampling algorithm.
The old code is deleted, all its usages are now working with the new
method.

This commit also includes lots of minor changes and refactorings of the
related code. Small bugs and inconsistencies were fixed, some TODOs left.
Object names, documentation and comments should also be clearer now.

This is important so passthrough NITTA logger continues to operate and:
1) makes seeing NITTA logs easiser when debugging;
2) helps to avoid NITTA process locking since pipe's buffer is full.

- Reconfigured everything using basic version of pyproject.toml
- Kept requirements.txt until poetry introduction
- Found better VS code extensions, configured them
  (I should share this workspace config!)
- Added type stubs for mypy's static type checking of some used libs
- Fixed errors found by mypy (nothing useful so far...)
- Fixed errors found by Ruff (cool!)
- Fixed all formatting errors in the code with black

Current suggested tooling:
- VS code for editing and syntax highlighting
- Microsoft's Pylance for an enchanced Python developer experience
  (syntax checking, autoimport, autocomplete, type inference, etc.)
- black (formatting)
- mypy (static type checking)
- ruff (flake8-like style+idioms linting, imports optimizing/sorting)
- Pytest for testing
- Test Explorer UI VS code extension + Python adapter (tests summary)
  ^ replace with Pylance's?

Refactored and fixed 2 scripts here:
- train_evaluate_in_ci
- crawl_data_by_tree_sampling_many

Logic of crawling from many examples was taken from the 2nd script,
generalized and reused in the 1st.

Relevant refactorings were made.