[Bug #20719] `Float` argument must be ASCII compatible

* Fix floor when ndigits is large

[Bug #20654]

This commit fixes Integer#floor and Float#floor when the number is
negative and ndigits is large such that 10**ndigits is a bignum.

Previously, it would return 0 in such cases. However, this would cause
unexpected behaviour such as:

    puts -1.floor(-5) # => -100000
    puts -1.floor(-10) # => -10000000000
    puts -1.floor(-20) # => 0

This commit changes the last result so that it will return
-100000000000000000000.

* Fix ceil when ndigits is large

[Bug #20654]

This commit fixes Integer#ceil and Float#ceil when the number is
negative and ndigits is large such that 10**ndigits is a bignum.

Previously, it would return 0 in such cases. However, this would cause
unexpected behaviour such as:

    puts 1.ceil(-5) # => 100000
    puts 1.ceil(-10) # => 10000000000
    puts 1.ceil(-20) # => 0

This commit changes the last result so that it will return
100000000000000000000.

Also `Numeric#remainder` should.

Handle the integert and the float parts separately in round_half_even
to prevent error occursions in floating point calculation.

https://hackerone.com/reports/1248108

The previous behavior depending on exact float values, it seemed
to work OK on amd64 and i386, but other CI platforms are
experiencing non-deterministic test failures with it. Relax test
slightly to hopefully pass on such platforms.

Depending on the float range, there could be an off-by-one error,
where the last result that should be in the range was missed. Fix
this by checking if the computed value for the expected value
outside the range is still inside the range, and if so, increment
the step size.

Fixes [Bug #16612]

It would be a relatively frequent case.  It is still slower than
`float * float` because `*` has a dedicated VM instruction (opt_mult),
though.