mruby-compiler: read a float literal as Integer 0 under MRB_NO_FLOAT - #7241

The Prism codegen compiles `case PM_FLOAT_NODE` out under `MRC_NO_FLOAT`
(which `MRB_NO_FLOAT` maps to), so a float literal falls through to the
default arm and the compile dies with an internal message:

    $ MRUBY_CONFIG=host-nofloat rake test
    ...
          MRBC mrbgems/mruby-compar-ext/test/compar.rb
    mrbgems/mruby-compar-ext/test/compar.rb:4: Not implemented: PM_FLOAT_NODE
    mrbgems/mruby-compar-ext/test/compar.rb:0:0: generator error, Not implemented: PM_FLOAT_NODE
    rake aborted!

The lrama parser read the same literal as the Integer 0 and warned
`floating-point numbers are not supported` (parse.y, since acdc2d1
added `MRB_WITHOUT_FLOAT`), and the shared test files are written
against that: 73 `Object.const_defined?(:Float)` guards in test/t and
the gem tests skip the Float rows at run time, which only helps once
the file has compiled.  13 of the 43 files in test/t hold a float
literal (374 of them, 226 in float.rb alone) and 21 gem test files do,
so with the arm missing no `MRB_NO_FLOAT` build can run the suite at
all: the compile stops at the first such file (compar.rb:4 for
build_config/host-nofloat.rb, test/t/array.rb:65 for
build_config/no-float.rb).

Give `PM_FLOAT_NODE` a `MRC_NO_FLOAT` arm that appends a generator
warning at the literal and emits the Integer 0, what the lrama parser
produced.  The warning goes through the diagnostic list like the Prism
parser warnings, so `mrbc -v` and mirb report it:

    $ printf 'p 1.5\n' > f.rb
    $ build/host-nofloat/bin/mrbc -v -o f.mrb f.rb
    ...
    f.rb:1:3: generator warning, floating-point numbers are not supported
    $ build/host-nofloat/bin/mruby -b f.mrb
    0

With the suite compiling again, five rows fail at run time because
their Float is now 0 instead of the row being skipped: the four Float
arena tests in test/t/gc.rb (`OP_DIV` divides by zero, `OP_ADDI` sees
-1), `Integer.__ensure(1.9)`, `defined?(Float::INFINITY)` and
`"a,b".split(/,/, 1.5)`.  Guard them the way the rows around them are
guarded.  `rake test` per config:

    host-nofloat.rb  before: rake aborted at compar.rb:4
                     after:  1738 OK, 0 KO, 0 Crash, 94 Skip; bintest 67 OK
    no-float.rb      before: rake aborted at test/t/array.rb:65
                     after:  674 OK, 0 KO, 0 Crash
    default          before and after: 2073 OK, 0 KO; bintest 111 OK

The comment in mruby-string-bitops/test/string_bitops.rb that described
a float literal as not compiling under `MRB_NO_FLOAT` is updated, and
doc/guides/mrbconf.md gains the sentence.

The previous commit reads a float literal as the Integer 0 under
`MRB_NO_FLOAT` and appends a generator warning at the literal, and the
only rows that exercised it were the ones it made skip.  Pin both:

- test/t/literals.rb asserts the value in-process, so every
  `MRB_NO_FLOAT` build that runs mrbtest checks it, cross builds
  without bintest included;
- the mrbc bintest compiles `p 1.5` with `mrbc -v`, asserts the
  warning with the literal's file, line and column, and runs the
  bytecode on the build's mruby for the `0`.  Whether the build has
  Float is asked of that mruby, since mrbc itself cannot be asked; a
  build without one skips.

Both skip in a Float build.  `rake test`, host-nofloat.rb: 1739 OK
(one more), bintest 68 OK (one more); default: both rows skip, totals
otherwise unchanged.