Use float instead of double for altitudeAngle and azimuthAngle

[jrandolf]

Altitude angle and azimuth angle are computed always computed with 32-bit floating points. For example, on macOS, https://developer.apple.com/documentation/appkit/nsevent/1534226-tilt. A priori, using a double doesn't offer more precision.

[EDIT: On macOS, the NSPoint struct uses double when the OS is x64, so this implies double accuracy is correct]

---

Preview | Diff

[mustaqahmed]

While I agree that the precision of a double may not be required here, it doesn't hurt either, right? I am worried about the non-zero compat risk that some code in the wild could break because of some assumption around precision.

More on that compat point: we picked double here from the TouchEvent spec (where it has been like this for years). What if there is a comparison between altitude angles of a TouchEvent and the corresponding PointerEvent that assumes an 8-digit precision (i.e. "finer" than float)? This may sound impossible but consider this closely related experience: we failed to even add precision to click coordinates because the integer coordinates of a MouseEvent was being compared with the coordinates of a click in a popular mapping website!

Without a clear benefit, we should avoid such a risk IMHO.