Fix python.args usage and optimize call_python

[guidanoli]

Closes #182

[scoder]

You say that this optimises the call. So, how much faster is it?

[guidanoli]

You say that this optimises the call. So, how much faster is it?

It's trivial to understand that this is faster. There's no need for benchmarking it.

Before:

• has to check every argument to see it is a _PyArguments instance (O(n))
• has to merge O(n) dictionaries
• has to concatenate O(n) tuples

After:

• has to check only the last argument (O(1))
• there is either 1 dictionary or none
• pre-allocates 1 tuple of right size for all positional arguments

[guidanoli]

See #187 for updated changes