The API doc of ptp_clock_adjust_speed says regarding the correction
parameter:
1. A call with @p correction set to `0` restores the nominal clock speed.
2. A call with a positive value for @p correction speeds the clock up
by `correction / (1 << 32)` (so up to ~50% for `INT32_MAX`).
3. A call with a negative value for @p correction slows the clock down by
`-correction / (1 << 32)` (so up to 50% for `INT32_MIN`).
So we need to divide by 2^32, not 2^32 - 1 (or `UINT32_MAX`).
