Suport upstream change in nuttx.

   CONFIG_DISABLE_ENVIRON can prevent task having to allocate
   memory for the env.

   Upstream changes defer the board reset based on the vaule of
   CONFIG_BOARD_RESET_ON_ASSERT
      0 (or not defined) Do not reset on assert
      >= 1 reset if assertion is in an interrupt handler or the idle task
      >= 2 on any asertion.

  up_assert is called from up_hardfault or an asertion failure.
  Part 1 of up_assert will call out to the board_crashdump
  Part 2 on return from board_crashdump will then perform
  the reset.

  board_crashdump needs a chunk of ram to save the complete
  context in. It uses  &_sdata which is the lowest memory and
  it will corrupt that memeory.  We can therfore can not allow
  return to the OS, as it could depend on that area of RAM.

  So all boards need to do a reset at the end of board_crashdump

   The F1 series GTIMs to not have GTIM_CCER_CC1NP.
   This bug fix tracks the upstream change that
   made GTIM_CCER_CC1NP conditional on the chip.

system_state.fmu_data_received_time can be set from an IRQ handler, thus
we need to ensure every read access to it in mixer_tick is atomic.
So we read it once and copy it into a local variable.

GCC did not do that.

This parameter was removed in d704a2d6 and should
not be referenced anymore.

    - jMAVSim in PX4/Firmware (3a7deefe): https://github.com/PX4/jMAVSim/commit/5150bbab6ea3e13907d2ff9eadd2c293d2477d49
    - jMAVSim current upstream: https://github.com/PX4/jMAVSim/commit/00bdf6d92b6a8678f5b683a3ebca9e52e88cb5ce
    - Changes: https://github.com/PX4/jMAVSim/compare/5150bbab6ea3e13907d2ff9eadd2c293d2477d49...00bdf6d92b6a8678f5b683a3ebca9e52e88cb5ce

    00bdf6d 2019-01-21 Beat Küng - jMAVLib: update submodule

This is required as we might be in interrupt context on this bare-metal target.

This allows the compiler to optimize better without loosing any performance / accuracy.

This leads to less jitter in the benchmark

This reduces interrupt load significantly.

This reduces interrupt load significantly.

The actual implementation is not atomic, as the value on the application layer would be limited.

This ensures we get absolutely accurate timing.

This call rarely needs to be truly atomic and the involved CPU overhead in making it atomic was unnecessary and introduces a lot of IRQ jitter with no value-add. The call has been moved to be non-atomic and the codebase will be inspected and changed in follow-up commits for the few instances where it is truly needed.

 - starts requirements for PX4 modules (hrt, param, etc)

 - can be reenabled after #11256 is merged

This avoids casting errors. An alternative to `std::round` would be to
add 0.5 before casting for positive values and subtract 0.5 before
casting for negative values.