MAINT: use reduction axis arguments in one-norm estimation

[argriffing]

This is a more focused subset of #4095.

benchmark information:

 ====== ========== ============
               --              solver        
               ------ -----------------------
                 n      exact     onenormest 
               ====== ========== ============
                 2      1.50ms      4.84ms   
                 3      1.74ms     43.26ms   
                 5      1.77ms     47.59ms   
                 10     1.82ms     51.76ms   
                 30     3.05ms     66.18ms   
                100    23.12ms     143.12ms  
                300    470.47ms    357.08ms  
                500     2.07s      638.53ms  
                1000     null       1.48s    


 ====== ========== ============
               --              solver        
               ------ -----------------------
                 n      exact     onenormest 
               ====== ========== ============
                 2      1.62ms      4.90ms   
                 3      1.62ms     37.43ms   
                 5      1.77ms     38.97ms   
                 10     1.84ms     39.22ms   
                 30     2.98ms     41.59ms   
                100    23.49ms     57.12ms   
                300    460.60ms    126.43ms  
                500     2.02s      237.10ms  
                1000     null      708.25ms  
               ====== ========== ============

[argriffing]

I made this PR over three hours ago and TravisCI is just now getting around to running the tests...

[argriffing]

Maybe I will merge this soon... it consists of internal implementation details of code that I had originally written myself, so it would be boring to review.

[ewmoore]

Yes. This is fine. Let's just take it.