perf: each reference to a path table walks the tree again, and one walk is ~4x find at 400k files #1201

use_case_dossier_test.py (#1200) is red at 17x find on a 400k-file tree. Two separate costs, measured on that tree (launcher included, warm page cache):

1. Every reference to a path table walks the tree again. Three CTEs over ./ cost three walks. One walk per distinct path table per statement would take the article's dossier query from 11 s to under 4 s on this tree.
2. One walk is ~4x find at this size (1.4x at 50k files, Path-table glob walks the entire root instead of the glob's literal directory prefix #1199). find stats every file too, so the gap is on dirsql's side of the stat: row building and insertion, not the filesystem.

Bar from #1198: dirsql wall minus launcher startup <= 1.1x native. The test sizes its corpus until native takes a second, so it holds the same ratio on any machine.

Done when

packages/python/tests/e2e/use_case_dossier_test.py passes with its pytest.mark.xfail(strict=True) marker removed in the same PR: identical rows, and dirsql wall time minus launcher startup within 1.1x the native pipeline. Part of epic #1205.