fix(plans): address review follow-ups from #3853

[Sayt-0]

Summary

Follow-up to #3853 that addresses review feedback around export safety, telemetry privacy, validation hooks, shared plan events, editor resolution, test placement, and Windows coverage.

Review feedback

Review point	Resolution
Windows locking was compile-only	Added a native windows-latest plan test job and made it gate image publication
Windows caller coverage was incomplete	Added cmd/root and pkg/tui to the Windows cross-build
Plans telemetry included user values	Limited telemetry to the subcommand and stable error codes
Existing PreRunE hooks could be replaced	Chained the existing hook and added regression tests
CLI export overwrite behavior was ambiguous	Existing destinations are refused by default; --force explicitly enables atomic replacement
Editor launch resolution was duplicated	Centralized command construction in pkg/tui/internal/editorname
Plan detail tests were hard to discover	Moved them alongside plan_detail.go
plan_changed was undocumented	Documented its payload and process-global visibility

Validation

Check	Result
task build	pass
task test	pass
task test-binary	pass
task lint	pass
task check-plan-cross	pass
go test -race -count=1 ./pkg/plans	pass
Workflow and documentation lint	pass
Native Windows plan tests	pending PR CI

Design notes

• Non-force export uses atomic hard-link publication after fully writing, syncing, closing, and securing a temporary file. Filesystems without hard-link support return a typed storage error rather than weakening the no-overwrite guarantee.
• plan_changed remains process-global because the shared notifier does not carry mutating-session identity and every open plans view must refresh. The API documentation now makes that visibility explicit.

[aheritier]

it's interesting @Sayt-0 as we said that we wanted to expand a bit our tests on windows to avoid some issues which were impacting only this OS. To see if we want to run only a few tests or all of them in the future