~2,000 film frames carry a FLYING_HEIGHT far too small, and three remedies fit equally

[NewGraphEnvironment]

Problem

fly#54 found the catalogue's FLYING_HEIGHT 10.764 times too large on 1,589 film frames and
repairs it. The same sweep found 1,963 film frames whose FLYING_HEIGHT is under half what
their SCALE x FOCAL_LENGTH implies (1,962 of them terrain-sampled; the figures below are over
those), and about 7,400 under 0.9. Those are deliberately
not repaired: fly_footprint(dem = ) marks them height_source = "implausible" and sizes
them from nominal scale.

They are not repaired because three different remedies each explain a comparable share and
the terrain cannot choose between them. Over MRDEM-30, of the 1,962:

remedy	frames it brings inside [1/1.6, 1.6]	reading
x 10.764	726	the fly#54 slip, run the other way
x 10	729	a dropped digit in a height in feet — 609 m is 2,000 ft
x 2	799 (a different set of rolls)	a 153 mm lens catalogued as 305 — 519 of the 799 are catalogued at 305, but 280 are at 153, which this reading cannot explain

Nor does 10.764 sort this tail the way it sorts the upper one: it scatters the 1,962 from
-0.70 to 5.38, leaving 130 still below the band and 1,106 above it, where the forward repair
puts all 1,589 at 0.80-1.32 with nothing between 6.69 and 10.01. Examples: bc79122 640 m at 1:20000 / 305 mm,
bc7584, bc7675, bc81013, bc77032; and at 0.500 bc80001, bcc162, bc79209.
bc7280 reads 60 m and bc5598 2 m, which nothing explains.

The mirror of the x2 case is on the high side too: a mass centred on r = 2, where 209 of the
223 frames sampled beyond r 1.8 are catalogued at 153 mm — a 305 recorded as 153.

Why it is worth an issue

For these frames nominal scale is the right fallback for the lens-mislabel cases (it never
reads focal_length) and an unknown quantity for the rest. Nothing is lost today — most were
already falling back, since 588 of the 1,962 have terrain above the recorded height — but
they cannot get the ~14% terrain correction until the height can be trusted.

What would settle it

Something independent of FLYING_HEIGHT, SCALE and FOCAL_LENGTH:

• the flight log (flight_log_url) where one exists, which states the planned altitude;
• adjacent-frame spacing along a leg against the footprint width each remedy implies —
  forward overlap is designed to about 60%, so a remedy that implies 5% or 95% is wrong.
  fly_bearing() already identifies contiguous legs;
• for the lens mislabel specifically, the calibration report or camera named on the roll.

Evidence

inst/extdata/flying_height_sweep.csv (set == "lower_tail" and "near_upper"),
data-raw/height_calibrate-flying_height_slip.R, and the "held against the scale" section
of inst/notes/terrain-correction.md.