feature: end-to-end agent workflow (#120)

* fix: broken mlflow unit test

* refactor: move mls agent to plexe/agents

* refactor: remove 'constraints' as we don't use this

* refactor: remove plexe exceptions as we don't use them

* fix: remove references to constraints

* refactor: move cot callback to internal

* refactor: move agent tools to plexe/tools package

* docs: update docs to reflect refactoring

* feat: add checkpointing and resumption

* feat: mlflow logging with parent runs for resumption

* fix: checkpointing disabled by default

* fix: ensure mlflow exp is active

* refactor: move mlops_engineer to plexe/agents

* refactor: put all agents under orchestrator agent

* fix: remove schema and eda from pre-build steps

* docs: update multi-agent-system.md

* fix: remove unnecessary todos from models.py

* fix: input schema not extracted after run

* fix: dangling instruction in schema_resolver_prompt_templates.yaml

* fix: eda agent missing from orchestrator

* fix: mlflow callback labels every build failed

* fix: dataset analyser has no prompt template

* fix: remove custom run names from mlflow child runs

* fix: mlflow logging failing because of safe_get

* chore: remove unnecessary comments

* fix: prompt templates to improve agent behaviour

* refactor: move object registry to plexe/core

* feat: add resumption to model build

* feat: add feature engineering agent

* feat: introduce immutability for object registry

* fix: tell eda agent to stop creating new datasets

* fix: allow required imports for feature engineer

* fix: clarify dataset analyser task

* fix: clarify feature engineer task

* fix: incorrect registry import in dataset splitter

* fix: miscellaneous fixes to dataset splitting and schema resolution

* fix: miscellaneous fixes to feature engineer

* feat: save dataset splitting code to bundle

* fix: eda report retrieving not always working

* fix: ensure manager doesn't try to do work itself

* fix: dataset splitting code registered incorrectly

* fix: switch feature engineer to use ml engineer model

* fix: add statsmodels to data analyser and feature engineer

* chore: update packages to upgrade smolagents

* chore: bump to 0.22.0

* feat: add model testing agent

* feat: add model testing agent

* refactor: switch to yaml for model archive

* fix: clean up example scripts

* fix: disable resumption for now

* fix: remove unnecessary unit tests

* fix: remove unnecessary unit tests

Author: marcellodebernardi

CLAUDE.md

@@ -11,7 +11,7 @@ The core architecture is as follows: agents go in `plexe/agents/*`, tools in `pl
 - `plexe/models.py`: Core `Model` class with build/predict functionality
 - `plexe/agents/schema_resolver.py`: Agent inferring input/output schemas
 - `plexe/internal/agents.py`: Multi-agent system implementation (`PlexeAgent` class)
-- `plexe/internal/models/tools/`: Tools for code generation, execution, validation
+- `plexe/tools/`: Tools for code generation, execution, validation
 - `plexe/config.py`: Configuration management and prompt templates
 - `plexe/internal/common/registries/objects.py`: Shared object registry for agents
 - `plexe/datasets.py`: Dataset handling and synthetic data generation