Separate inconsistent_or_missing_returns_checker into two checkers

[Raine-Yang-UofT]

The inconsistent_or_missing_returns_checker handles two error messages inconsistent-returns and missing-return-statement as the logic to detect the two errors are similar. However, for readability and expandability concern, it could be a better design to separate this checker into two, one specifically for each error message.

For example, during the integration of z3 option into this checker, we only want missing-return-statement to be affected by z3 option. However, as the checking for two messages are tightly coupled, we have to implement a messy and suboptimal solution in our current codebase:

# At Line 85
# ignore unfeasible edges for missing return if z3 option is on
if self.linter.config.z3 and (
    not block.is_feasible
    or not any(
        edge.is_feasible for edge in block.successors if edge.target is end
   )
):

We could have directly filters out blocks whose edge connecting to the end node is unfeasible, which replaces statement or not any(edge.is_feasible for edge in block.successors if edge.target is end) with the following modification:

Change

# At Line 60
# get the blocks connected to the end of cfg
end = node.cfg.end
end_blocks = [edge.source for edge in end.predecessors]

to

# At Line 60
# get the blocks connected to the end of cfg
end = node.cfg.end
end_blocks = [edge.source for edge in end.predecessors if edge.is_feasible]

However, this change would cause errors in inconsistent-returns in our current checker.

In addition, in our current implementation, we need to maintain unit tests for z3 options for inconsistent-returns even though z3 is not used for this message. This is to ensure that modifications on missing-return-statement z3 option would also cause accidental changes to inconsistent-returns. Refactoring the checker would also us to remove the redundant test cases.