Actual FPS simplification & unit change [API change]

[maloel]

RS2_FRAME_METADATA_ACTUAL_FPS is now really the actual FPS, same as shown in the viewer as Hardware FPS, although with better precision:

• Changes the units of ACTUAL_FPS metadata to be *1000 (e.g., 30.1 fps will be represented as 30100 in metadata)
  • This is NOT backwards-compatible!
  • This greatly improves syncer accuracy at lower FPS (see Improve desyncs on different-FPS streams #12315)
  • NOTE: this likely has implications for rosbag recordings, in that the old values will not be *1000... when tested, they play just fine but show the old values
• Removes discrete mode for ACTUAL_FPS
  • This was used in non-color (i.e., Depth, IR, etc.) d400_device derivatives; actual FPS will now be correct
• Removes use of ACTUAL_EXPOSURE as predictor of actual FPS
• If ACTUAL_FPS cannot be calculated (missing previous frame time, or getting a value ~0) then ACTUAL_FPS value will be missing (i.e., supports() will return false; contrast with the previous behavior, which instead put the stream FPS there)
• Simplifies metadata parsers: joins the supports() and get() functions into single find()
• Likewise implements a single virtual frame_interface::find_metadata() to both check and get metadata - should save on performance, especially relevant in the syncer

Related to [RSDSO-19336]

[Nir-Az]

Can we run a "Semi" RealCI and see jo reggression.
We have to be carefull here.
Also need to make sure ROS is not using this field

[maloel]

No ACTUAL_FPS in realsense-ros

[maloel]

Two Semi runs of RealCI came back clean so far.

[Nir-Az]

LGTM

[maloel]

We see no regressions on external sync slave setups (which prompted this change):

• 30-30 FPS 3-second desync every 170 sec (before this PR) is now gone (with this PR)
• 60-60 behaves well
• D60-C5 no regression; still no sync at all due to huge spread between D & C arrival times
• D5-C60 sync happens as expected
  This may or may not be with a faulty external sync setup (currently being investigated).

[maloel]

RealCI on the D457 ran twice; no new behavior observed.