[uss_qualifier] introduce distance error tolerance for some DisplayProviderBehavior

[Shastick]

Add an error margin when checking for specific USS behavior around specific thresholds, such as the maximum diagonal of an observed area.

This updates the display provider maximum allowed area observation

(Updating other checks such as the clustering thresholds require some unrelated adaptations and will happen in a separate PR)

Notes

Some context from a Slack thread:

[...] uss_qualifier's spherical approximation method is not as accurate as Vincenty's ellipsoidal approximation, so it indeed requests too large of a diagonal by the 1.9 meters noted. This is a little unintuitive because uss_qualifier's spherical approximation is larger than the Vincenty ellipsoid, yet Vincenty produces a larger distance (and therefore we should expect uss_qualifier to overestimate distance, but instead it under estimates distance). My understanding is that this is because the circles of the same latitude are sometimes larger on the ellipsoid even though the overall ellipsoid is smaller.

[...] In general, when we're checking a limit X, we shouldn't send exactly X because systems differ in the computations and comparisons and therefore exact equality in floating points is never a good idea outside of special values like 0. Instead, we should send X-sigma or X*(1-sigma) to check that values less than X are accepted, where sigma is a constant describing how large of a computation difference is acceptable. In the case of distances, an LLM tells me the maximum distance error between uss_qualifier's less-correct spherical model and a more-correct ellipsoidal model like Vincenty is 0.674%. So, we should pick a diagonal of 7000*(1-0.00674) meters to verify that participants accept large but valid diagonals. (but probably round up a bit to account for other sources of computation difference, so perhaps use 0.7% instead)

[BenjaminPelletier]

0.007 is presumably the DISTANCE_ERROR_TOLERANCE_FRACTION that we want in order to achieve a DISTANCE_ERROR_TOLERANCE_PERCENT of 0.7

Suggested change

	DISTANCE_ERROR_TOLERANCE_PERCENT = 0.007
	DISTANCE_ERROR_TOLERANCE_FRACTION = 0.007

or

Suggested change

	DISTANCE_ERROR_TOLERANCE_PERCENT = 0.007
	DISTANCE_ERROR_TOLERANCE_PERCENT = 0.7

[BenjaminPelletier]

nit: it's not different spherical models, it's just different models for the shape of the Earth. Specifically, the more-accurate ellipsoidal model can be off from our less-accurate spherical model by a little under 0.7%.