implement setup.cfg, lint cleanup of Plex directory

Do not install test dependencies on CPython dev versions in travis as they are likely to fail in pre-releases.

Tighten grammar to disallow a mix of Py2 style octal literals with Py3 style underscores: "07" is still allowed, "0_7" is not and must be spelled "0o_7" and the like.
Basically, you get either Py2 style legacy octals or Py3 style underscore literals, but you have to decide what you want.

Only inject eval() locals/globals args when it's called as a builtin and not redefined e.g. in the module (found when updating test_grammar() and overriding it there).

When detecting an invalid system configuration for overflow checks, at least print a warning, instead of leaving a dangling live exception behind.

The current code fails to compile with:

<source>:3:54: error: invalid operands to binary expression ('enum E' and 'enum E')
const enum E neg_one_class_29 = (enum E) ((enum E) 0 - (enum E) 1);
                                          ~~~~~~~~~~ ^ ~~~~~~~~~~

This change reverts to the code that was before #2186 but silences erroneous GCC warning enabled by -Wconversion (which is not a part of -Wall).

Fixes #2749

This fixes Github issue #2819.

Fix support for C++ enum classes

warning C4244: 'initializing': conversion from 'double' to 'float', possible loss of data

Fix msvc compiler warning C4244

Pointer buf cannot be NULL

warning C4244: 'initializing': conversion from 'double' to 'float', possible loss of data

Clean up insertion points after use in prange code since they conflict with the deep-copying that with-statements use for their finally-clause.
Closes GH-2780.

Avoid useless "div by zero" checks in optimised constant division code. Also avoid using a (named) C macro where code generation will do.
See GH-2820.

Fix exception messages for float division by zero, as well as their tests (which were wrong in Py2.6).

Closes #2820.

Remove superfluous code