Add rntuple_test32 to CI

[linev]

Adding 32bit types tests,
taken from #411

Thanks to @undefined-panda

[undefined-panda]

I discovered that you can fix the excessive precision problem for 32 bit floats with this:

export function floatToHex(num, { field }) {
  if (num === 0) return (Object.is(num, -0) ? "-" : "") + "0x0p+0";
  
  // convert num to Float32 for correct precision
  if (field.startsWith("Float")) {
    const tmp = new DataView(new ArrayBuffer(4));
    tmp.setFloat32(0, num, false);
    num = tmp.getFloat32(0, false);
  }

  const buf = new ArrayBuffer(8); // 8 Byte == 64 Bit
  const view = new DataView(buf);
  view.setFloat64(0, num, false);

However, for that it is required to know if num is meant to be a float or double, for example by the name of the field.

[linev]

You are free to create PR if you think that this improve testing