Fix missing _RP kind suffix on floating-point literals that are not binary-representable.

[chenyu76]

In Fortran, an unsuffixed literal like 0.1 is of type default real (usually 32-bit single precision). When RP is configured as 64-bit double precision, comparing or combining real(RP) variables with unsuffixed literals triggers an implicit type conversion. Worse, 0.1 in single precision and 0.1_RP in double precision are different binary approximations. This means a condition such as

if (rand() <= 0.1) then

may evaluate differently in rare cases.

This PR appends _RP to unsuffixed floating-point literals that is not exactly
representable in binary.

How They Were Detected

The missing suffixes were identified by fortran_float_number_extract, a simple tool that

1. parses Fortran source files into an AST using fortran-src,
2. traverses the AST to find all real/integer literals without a kind suffix,
3. checks whether those literals can be exactly represented in binary form and reports them if not.

After identification, I manually checked whether each literal needed to be rewritten.

Literals that were skipped

• The literal 0.9 in /fortran/tests/testsuite/bigprob.f90:23:43.
• The original code by Prof. Powell in /fortran/original/.