Always type-infer floating point types

[dagss]

Depends on #236.

Since Python uses double for its float and complex types and all important operations on it can map directly to C space, and since Python<->C conversions are 1:1, we can safely turn on type-inference in this case without breaking anything.

This requires that #236 is implemented first though so that any use of attributes or methods goes through.

Note that, at least if typeinference is turned off, this means that

cdef float x = 3.2
y = x # y will be typed as double,
      # so as to behave as if y was a Python
      # object

Unresolved issue: If typeinference is turned on, perhaps y should be typed as float instead.

[scoder]

This is implemented as part of the safe type inference mode.