[typing] uniformize a bit more Is_a_real and Not_a_number

Fonenantsoa Maurica authored 6 years ago and Julien Signoles committed 5 years ago

  - Interval:
    - Not_an_integer -> Is_a_real|Not_a_number.
    - infer -> infer_with_real
      because Not_a_number has priority over Is_a_real
    - Ival.bottom when Is_a_real
      because the interval is for integers
  - Typing:
    - integer_ty -> number_ty=Cty|Gmpz|Libr|Nan
    - Other -> 'Other (Oreal|Onan)'
  - Translate:
    Encode real constants into strings:
      - **NO** conversion to float-point type
      - strnum_ty=StrZ|StrR|Not_a_strnum
        for tracking the type of the string
Add Libr:
  - Gmpz -> Gmp=Gmpz+Gmpq
  - Libr=ref to Gmpq
Add Gmpq builtins:
  - Custom mini-gmp -> libgmp
    because mini-gmp has no support for Q
  - Add arithmetic over Q in e_acsl_gmp_api.h
Misc:
  - dec_to_frac: decimal expansion to fractional representation
    because decimal expansion is interpreted as double by Gmpq
Tests:
  - BTS 1307 has an assertion wrongly evaluated, fixed
  - Add tests/gmp/reals.c
THE MOST IMPORTANT TODO:
  - Completely hide the library for numbers (currently Gmp)
    inside Libr: Typing, Env and Translate should only know
    Libr and never directly call Gmp.
    This is crucial for extending E-ACSL in the future
    (eg: Gmp has no support for elementary functions), and
    also for using it as part of an abstract compiler
    (eg: with Fldlib)
    For the time being, in Translate,
    we have something as ugly as:
      let init_set =
        if Libr.is_t ty then Libr.init_set
        else Gmp.init_set

Replaces calls to deprecated functions related to Visitor_behavior

See merge request frama-c/e-acsl!295

Bugfix/issue549

See merge request frama-c/e-acsl!289

Conflicts:
	nix/frama-ci.nix

Update E-ACSL reference manual

See merge request frama-c/e-acsl!293

Update Frama-CI data

See merge request frama-c/e-acsl!294

fix bug when combining -e-acsl-instrument and -variadic-no-translation (fix issue #88)

See merge request frama-c/e-acsl!292