This ensures that all files are tentatively removed.

Python provides better tools that exceptions for that.

If the opening of the cache file fails, print an error
message to know where to look at.

The hash doesn't match what is done in beat/beat.web> resulting
in a mismatch when using e.g. beat/beat.cmdline>.

This patch modifies the code so that the returned value
follows the same logic as used in beat/beat.web>

This allows the evaluator to properly write to its own
output as the method is called each time a processor
output is written to.

Fixes #27

Currently there's nothing in the log so if there's an error,
the result will be a missing file error.

The None value is not properly handled and ends up
triggering a warning while it is a valid value for
end_data_index.

This patch fixes this.

Under some circumstances the decoding may fail as the binary
data may contain values outside the utf-8 "visible" range.

Fixes #26

This change also updates the valid check to print the
error if any should appear making the failure easy to
understand.

Renaming the algorithm has the added benefit of making
the prefix content clearer to understand.

WARNING this is a breaking change

xxx_loop and xxx_loop_user were not really clear.

They are now:
- xxx_loop_evaluator
- xxx_loop_processor

This makes their role easier to understand.

Now requires a sequential loop algorithm.

This method will trigger the reading of the input
by the loop algorithm.