Fix RSKGES bug and LU-based range stabilization in sparse QRCP example

[rileyjmurray]

Resolves #135. Along the way, fixes an apparently longstanding bug in RSKGES, where we missed an early-return statement.

Here's example output after making the changes in this PR. The build was in debug mode.

(base) Rileys-MacBook-Air:build-randblas-examples riley$  ./slra_qrcp 4 /tmp/N_reactome/N_reactome.mtx

Processing matrix in /tmp/N_reactome/N_reactome.mtx
n_rows  : 10204
n_cols  : 16559
density : 0.00025931532

Computing rank-4 truncated QRCP.
Internally use 2 steps of power iteration.
Consider four runs, each stabilizing power iteration in a different way.

Take Q from LQ
Row norms for Yf : [ 55.646701, 45.682935, 50.159852, 41.275255 ] 
Row norms for R  : [ 59.044092, 59.051427, 59.082688, 59.128318 ] 
Leading pivots   : [ 7914, 648, 962, 7588 ]
Runtime in μs    : 13962

Sketch-orthogonalize
Row norms for Yf : [ 23.080879, 48.230123, 37.339738, 27.324248 ] 
Row norms for R  : [ 59.044092, 59.051427, 59.082688, 59.128318 ] 
Leading pivots   : [ 7914, 648, 962, 7588 ]
Runtime in μs    : 32400

Do nothing. This is numerically dangerous unless power_iter_steps is extremely small.
Row norms for Yf : [ 314192.79, 246510.58, 406939.74, 476108.1 ] 
Row norms for R  : [ 59.044092, 59.075882, 59.105465, 59.149568 ] 
Leading pivots   : [ 7914, 649, 962, 7588 ]
Runtime in μs    : 8519

Take L^T P from row-pivoted LU of W^T.
Row norms for Yf : [ 74.624896, 81.881853, 63.740244, 51.477882 ] 
Row norms for R  : [ 59.044092, 59.051427, 59.082688, 59.128318 ] 
Leading pivots   : [ 648, 7914, 962, 7588 ]
Runtime in μs    : 8864