Frequents calls to list.remove() were being a bottleneck for large document lists (100k+). Using a set make remove() calls much faster. This commit has a corresponding server patch in the core ir.attachment search method. bzr revid: odo@openerp.com-20120821131920-j653hvmff78fxxtr
To be done: ----------- * Test to not create several times the same file / directory -> May be put a sql_constraints uniq on several files -> test through remove or put * Retest everything