pass1b.c (pass1b, process_pass1b_block): Change the num_bad

pass1b.c (pass1b, process_pass1b_block): Change the num_bad
	field calculation so that it only counts EA block entries
	as a single multiply claimed block (since once we clone
	the EA blocks for one inode, we fix the problem for all of
	the other inodes).  Also, I moved the num_bad calculation
	from process_pass1b_block to the end of pass1b.  This
	fixes a *significant* performance bug in pass1b which hit
	people who had to had a lot of multiply claimed blocks.
	(Can you say O(n**3) boys and girls?  I knew you could...
	Fortunately, this case didn't happen that much in actual practice.)