The current VME bus error handler adds errors to the bridge error list.
vme_master_{read,write} then traverses that list to look for relevant
errors.
Such scheme didn't work well for accesses going through vme_master_mmap
because they would also allocate a vme_bus_error, but have no way to do
vme_clear_errors call to free that memory.
This changes the error handling process to be other way around: now
vme_master_{read,write} defines a window in VME address space that will
catch possible errors. VME bus error interrupt only traverses these
windows and marks those that had errors in them.