autocorrelation pair count normalization

[aphearin]

As pointed out by @Christopher-Bradshaw in #962 (and also manodeep/Corrfunc#201), the angular_tpcf function appears to have an incorrect factor of Nr x Nr instead of Nr x (Nr-1). This has a negligible impact on all practical cases of interest, but the error is needless and should still be fixed. The main question concerns whether other mock_observables functions are impacted as well, or whether this is isolated to angular_tpcf.