Force saner ua_os in flow_metadata and activity_events tables

[philbooth]

The values for ua_os in our metrics tables come from common_log_format.normalize_user_agent in mozilla-services/lua_sandbox_extensions, where they're defined in ua_os_matchers like so:

local ua_os_matchers = {
    -- search, replace
      {"iPod"                    ,"iPod"         }
    , {"iPad"                    ,"iPad"         }
    , {"iPhone"                  ,"iPhone"       }
    , {"Android"                 ,"Android"      }
    , {"BlackBerry"              ,"BlackBerry"   }
    , {"Linux"                   ,"Linux"        }
    , {"Macintosh"               ,"Macintosh"    }
    , {"Mozilla/5.0 (Mobile;"    ,"FirefoxOS"    }
    --http://en.wikipedia.org/wiki/Microsoft_Windows#Timeline_of_releases
    , {"Windows NT 10.0"         ,"Windows 10"   }
    , {"Windows NT 6.3"          ,"Windows 8.1"  }
    , {"Windows NT 6.2"          ,"Windows 8"    }
    , {"Windows NT 6.1"          ,"Windows 7"    }
    , {"Windows NT 6.0"          ,"Windows Vista"}
    , {"Windows NT 5.1"          ,"Windows XP"   }
    , {"Windows NT 5.0"          ,"Windows 2000" }
}

We could simplify our queries if we updated the iPad/iPhone/iPod ones to iOS and all the Windows ones to Windows. This could be done in the update parts of the respective import scripts, with a one-off query to update all the historical data.

[rfk]

/cc @davismtl; would you worry if we lost the distinction between e.g. "Windows 10" and "Windows 8" in these metrics?

[davismtl]

Yeah, I would a bit. For example, onboarding team is running tests on just Win 10 since it has become largest population. It's nice to isolate it.

Question is... could we do something like?
OS: Windows
Version: 10
Device type: PC

or

OS: iOS
Version: 9.2
Device type: iPhone

I think that as a result we could simplify most queries but if we wanted the granularity, it would be there. (thinking out loud)

Thoughts?

[philbooth]

Closing this, based on the Alex's last comment.