Float literal syntax spec and tests don't match

[AndrewScheidecker]

The spec defines the syntax for fN to require a decimal point or exponent, and defines f(32|64).const to only accept a fN as their immediate operand.

However, there are many tests that use f(32|64).const without a decimal point or exponent; e.g.:

spec/test/core/float_literals.wast

Line 82 in d22a765

(func (export "f32-dec-sep1") (result f32) (f32.const 1_000_000))

[rossberg]

Indeed, PTAL #1069 for a fix.