Fix constant generation for C++ floats

[clalancette]

While working on a package, I came across a build warning on Windows that said:

warning C4305: 'initializing': truncation from 'double' to 'const float'

After looking into it, I realized that the problem was that when we generate float constants for C++, the value we assign to it is a double, not a float. For instance, given the Constants.msg file, we currently generate (truncated for brevity):

  // constant declarations
  static constexpr float FLOAT32_CONST =
    1.125;
  static constexpr double FLOAT64_CONST =
    1.125;

But that is not quite right; that first line should be:

  static constexpr float FLOAT32_CONST =
    1.125f;

Interestingly, the warning only shows up when assigning certain constants to the float, like 0.05.

So this PR does three things:

1. Does a small refactoring of the tests, removing bits we don't need.
2. Adds a test for the problematic constants (by default, the warning will only show up on Windows).
3. Fixes the code generation to append the f as appropriate.

[clalancette]

CI:

• Linux
• Linux-aarch64
• Windows