Keep the plot origin when a render is held past source()

[midleman]

Fixes #1334.

Summary

A plot created inside source() could lose its origin, leaving the Plots pane with no source file and no way to recover it.

• hook_mode() only captured the origin on the has_changes false->true edge, so a new page starting while earlier changes were pending never re-captured after new_positron_page() cleared the snapshot
• graphics_on_did_execute_request() cleared the snapshot at the end of every request, so a render held into a later request lost it
• Now captures whenever no snapshot is pending, and drops the end-of-request clear as redundant with the new-page clear
• Adds test_plot_origin_survives_hold_across_source, which fails on main inside the assertion and passes with the fix

E2E Triage Diagnosis

🟢 High confidence -- Plot origin is lost on the source() path when the origin snapshot is dropped before process_changes consumes it, leaving the fallback to run after source() popped the context stack

• Test: Plot File Attribution > R - Plot origin shows source file after run file command
• Targeted failure: Error: expect(locator).toBeVisible() failed
• Signal: Observed via file probes in a RED ark integration test: push_source_context -> EAGER CAPTURE succeeds -> new_positron_page clear wipes it -> hook_mode skips re-capture (old_has_changes=true) -> end-of-request clear in graphics_on_did_execute_request drops it again -> pop_source_context -> take_pending_origin FALLBACK -> None. The end-of-request clear is the decisive path for a render held across requests. Plot renders and the metadata RPC succeeds, but .plot-origin-file never enters the DOM. Local e2e did NOT reproduce (10/10 green on macOS, origin consumed inside the same request), so the CI route is predicted, not observed.
• Frequency: 32/137 runs (23.4%) on main, ubuntu/electron
• Hypothesis: race

[lionel-]

I've pushed an update to make sure a stale plot origin isn't mistakenly applied to a new plot, with a test.

However AFAIU I don't think this will fix #1334. All the plot hooks and callbacks are called deterministically on a single thread, and the dev.hold() in plot() (the E2E test calls plot()) is flushed on exit: https://github.com/r-devel/r-svn/blob/0017c9e54f90bdc4a10dd1f41a7afdf7cac8ac0e/src/library/graphics/R/plot.R#L79.

Still a good fix to have, even though it only fixes an unlikely case.