[autoparallel] refactored the autoparallel module for organization

[FrankLeeeee]

What is the problem?

The previous code is kind-of mixed between old and new versions. Therefore, a cleanup is required to well organize the module structure.

What does this PR do?

This PR organizes the new version code into colossalai.auto_parallel.tensor_shard according to the structure below:

- tensor_shard
  - deprecated
  - solver
    - solver
    - graph_analyser
    - strategy_constructor
    - cost_graph
    - options
  - node_handler
    - strategy
  - sharding_strategy
  - runtime
  - constants
  - utils

The unit tests have been refactored accordingly.

Test Results

With AUTO_PARALLEL=1, the result is shown below. The next step is to fix these faulty but skipped test cases (which have been broken for quite a while).