get rid of _native_coercion_ranks_inv and _native_coercion_ranks

[vdelecroix]

In the file sage/structure/coerce.pyx there is an homemade coercion for python numeric objects (i.e. bool -> int -> long -> float -> complex). Instead we should use the ready made python function coerce. Or even use the ​PyNumber_Coerce function from the C API.

[vdelecroix]

Last 10 new commits:

​a60134c	trac #17740: review 1 (documentation)
​728811d	trac #17740: review 2 (clean Errors)
​96c1a03	trac #17740: review 3 (less in try/except block)
​2cb51c0	Re-introduce action of fraction field as fallback for division action.
​bff474b	Better _pseudo_fraction_field default implementation.
​9c970ae	trac #17740: merge sage-6.6.beta1
​2075e2e	trac #17740: avoid parent deaths
​4fcee82	Merge commit '2075e2e' into t/17800/ticket/17800
​7083f19	Remove use of PY_IS_NUMERIC
​1ca0e2b	trac #17685: native coercions with PyNumber_Coerce

[vdelecroix]

The new solution looks twice faster from Python

Old timings:

sage: cm = sage.structure.element.get_coercion_model()
sage: %timeit cm.canonical_coercion(1r,2.r)
1000000 loops, best of 3: 520 ns per loop
sage: %timeit cm.canonical_coercion(1r,complex(1r,1r))
1000000 loops, best of 3: 812 ns per loop

New timings

sage: cm = sage.structure.element.get_coercion_model()
sage: %timeit cm.canonical_coercion(1r,2.r)
1000000 loops, best of 3: 234 ns per loop
sage: %timeit cm.canonical_coercion(1r,complex(1r,1r))
1000000 loops, best of 3: 469 ns per loop

[vdelecroix]

conflicts with #17862 and there is a reference counting issue...

[vdelecroix]

Hello,

I propose to close this one as duplicate because I found a much better solution in #18076.

Vincent