In CPython the function form is twice as slow.

This eliminates an extra allocation for the backing array by storing it in the
memory directly following the tuple itself.

This patch adds minimal support for complex numbers.
Complex literals can be parsed and a very minimal
complex number type has been added to the runtime.
While minimal, the basic pieces are fleshed out enough
that producing follow-on patches to incrementally
develop `complex` should be easy.

Method covered by new benchmarks:
* Dict.GetItem
* Dict.IterItems
* Dict.IterKeys
* Dict.IterValues
* NewInt - Fixed so that references aren't elided anymore.
* NewStr

As specified here:

* https://docs.python.org/2/library/stdtypes.html#sequence-types-str-unicode-list-tuple-bytearray-buffer-xrange

This commit only uses the sequence count method for `list.count`,
but it can easily be applied to other sequence types in follow-on
commits.

Slight punctuation and hyphenation improvements, with additional styling on command and directory names.

In the course of reviewing #116 it came to light that
some of the other `str` methods that involve changing
case do not handle multi-byte characters correctly.
This patch fixes that.

Replaced FunctionArg with Param and moved arg validation logic out of
Code and into new ParamSpec type. ParamSpec can then be made more
general purpose and used for builtin functions.

Rather than allocating a frame on every call, we keep an already allocated list
of frames ready to go. This trades lightly more memory usage for faster calls.

Also updated Makefile to include these tests in the "make test" suite.
It became awkward to automatically detect the .py test files (they don't
really follow any particular naming convention) so they're listed
explicitly. To make things easier to follow I also listed out the other
stdlib test files explicitly. Long term I think we should be listing all
files this way (including .go sources) so that things are less magical.