Avoid some pollution of CLI argument namespace across vtgate and vttablet

[aquarapid]

This does not fix all instances of this problem, but it does improve the situation:

Before:

$ bin/vtgate -h 2>&1 | grep '  -' | wc -l
343
$ bin/vttablet -h 2>&1 | grep '  -' | wc -l
491

After:

$ bin/vtgate -h 2>&1 | grep '  -' | wc -l
196
$ bin/vttablet -h 2>&1 | grep '  -' | wc -l
484

The vttablet situation is still not great, but the remaining vttablet flags that are really vtgate-only are not easy to remove, e.g. the ones from go/mysql, like:

  -mysql_auth_server_static_file string
  -mysql_auth_server_static_string string
  -mysql_auth_static_reload_interval duration
  -mysql_clientcert_auth_method string

which is not easy to break into sub-packages without import cycles; and will need greater surgery.

[ajm188]

I know this is still in draft, but in case it's helpful, I wonder if it would be better to move the flag definitions into subpackages rather than the other way around? For example, what I'm starting to do here: f556f7d

[aquarapid]

I know this is still in draft, but in case it's helpful, I wonder if it would be better to move the flag definitions into subpackages rather than the other way around? For example, what I'm starting to do here: f556f7d

Yeah, that probably fair enough. But more surgery than I'm willing to commit to right now.

[aquarapid]

For release notes:

It has been possible for several Vitess versions to pass certain vttablet flags to vtgate and certain vtgate flags to vttablet.  This did not have any effect on the operation of vtgate or vttablet, but did confuse users when looking at the commandline help.  This change fixes some instances of this problem.  As a result, users might have vtgate and vttablet startup scripts that use flags that will no longer be accepted, and vtgate or vttablet startup will fail.  Please test your vtgate and vttablet startup with the flags you use before rolling this version out in production."