RPC transformer: RPCInverseTransformPoint() doesn't check convergence

[Even Rouault]

RPCInverseTransformPoint() doesn't check that the guessed (long, lat) has converged within the tolerance and thus the transformer can return a completely wrong result.

Furthermore during the iterations, it doesn't take into account the DEM elevation, only in a post processing stage. We might get better convergence by updating it at each iteration.

[Even Rouault]

trunk r31087 "RPC: make RPCInverseTransformPoint() check convergence has been reached (#6162)"

[Even Rouault]

trunk r31097 "RPC: make RPCInverseTransformPoint() finally work on some RPCs, whose convergence is very sensitive to using valid height values, by using DEM information (#6162)"

[Even Rouault]

trunk r31117 "RPCInverseTransformPoint(): do not immediately check error decrease to early stop iterations if the DEM elevation has just been adjusted. Avoid false convergence errors. #6162"

[Even Rouault]

trunk r33636 "RPC: add RPC_PIXEL_ERROR_THRESHOLD and RPC_MAX_ITERATIONS transformer option to control behaviour of inverse transformer; in presence of a DEM, when going below the error threshold, make inverse transformer do a DEM adjustment before returning if not done during the current iteration; do not return the computed long,lat of the current iteration since it might not respect the error threshold (#6162)"

[Even Rouault]

trunk r33846 "RPC inverse transformer: another round of attempts to improve convergence in some cases (#6162), add RPC_INVERSE_VERBOSE and RPC_INVERSE_LOG configuration option"

[Even Rouault]

trunk r33871 "Test GDAL inverse RPC transformer with oscillations situation (#6162)"