MSI: First go at thread safety for MSI records

[Robert Shearman]

Mike McCormack wrote:

> Alexandre Julliard wrote:
>
>> Do you really need a separate critical section for every single MSI
>> object?  That seems a bit overkill to me.
>
>
> What do you suggest?
>

Keep the msiobj_lock/msiobj_unlock functions, but make them use a 
process-wide critical section. It will then be simple to add a per 
object critical section if it is proved that performance is suffering.

Rob