Use preadv/pwritev to efficiently handle multiple mappings at once.

[luqmana]

Illumos, like linux, has vectored versions of pread/pwrite which we can use here instead of calling pread/pwrite in a loop. Using preadv/pwritev saves us a couple of roundtrips.

In fact, with #39 we can see the improvement in latency just simply booting/shutting down a guest off an file-backed nvme storage:

Before:

  write (us)                                            
           value  ------------- Distribution ------------- count    
               8 |                                         0        
              16 |@@@@@@@                                  25       
              32 |@@@@@@                                   21       
              64 |@@@@@@@@@@@@@@@@@@@@@@                   73       
             128 |@@@@                                     13       
             256 |@                                        3        
             512 |                                         0        

  read (us)                                            
           value  ------------- Distribution ------------- count    
               4 |                                         0        
               8 |@@@@@                                    197      
              16 |@@@@@@@@@@@                              405      
              32 |@@@@@@@@@@@@@@@@@@@                      713      
              64 |@@                                       96       
             128 |@@                                       77       
             256 |@                                        39       
             512 |                                         11       
            1024 |                                         1        
            2048 |                                         0        
            4096 |                                         2        
            8192 |                                         0

After:

  write (us)                                            
           value  ------------- Distribution ------------- count    
               8 |                                         0        
              16 |@@@@@@@@@@@@@@@@@@@@@@@@@@               89       
              32 |@@@@@@                                   20       
              64 |@@@@@@@                                  24       
             128 |@                                        3        
             256 |                                         0        

  read (us)                                             
           value  ------------- Distribution ------------- count    
               4 |                                         0        
               8 |@@@@@@@@@@@@@@@                          570      
              16 |@@@@@@@@@@@@@@@@@                        659      
              32 |@@@@@                                    180      
              64 |@@@                                      105      
             128 |                                         19       
             256 |                                         6        
             512 |                                         1        
            1024 |                                         1        
            2048 |                                         0