test: retain bounded browser launch evidence on failure

docs/validation.mdmodified+20-0

Changes to docs/validation.md

		@@ -1,5 +1,25 @@
1	1	# Validation evidence
2	2
	3	+## Browser qualification diagnostics
	4	+
	5	+Browser fixtures keep their failed private directory after owned process-group
	6	+cleanup. It contains native fixture credentials, database state and the browser
	7	+profile, so CI does not upload that directory. Local retained paths are reported;
	8	+CI runner disposal still ends access to private files that were not exported.
	9	+
	10	+CI exports a closed launch-child observation and at most 64 KiB of startup stderr.
	11	+Public stderr capture stops before any page visit can receive credentials. Later
	12	+browser stderr remains in the private fixture, while exported metadata reports
	13	+that capture was withheld. Refused evidence destinations report unavailable and
	14	+preserve the original browser failure. Launch exit observation, signal and cleanup
	15	+intent are separate fields; they do not prove browser/page/thread liveness or the
	16	+kernel's exact exit time. Post-renewal CDP liveness has its own operation label.
	17	+
	18	+Discovery, transport, stream and outer-test limits remain unchanged. These
	19	+observations support the open investigations in [#44](https://github.com/stacksjs/drone/issues/44)
	20	+and [#64](https://github.com/stacksjs/drone/issues/64); retaining a new attempt does
	21	+not explain an earlier failure whose Chrome evidence was removed.
	22	+
3	23	## Owned mission simulation slice
4	24
5	25	- Full release suite: 211 Bun tests, 17,424 assertions, zero failures. Buddy lint, application/runtime/view