Fix plot fallback snapshots

[FBartos]

Summary

Fixes the ggplot visual snapshot fallback lifecycle and the slowdown caused by repeatedly falling back through structural snapshots.

New behavior

• Visual snapshots remain the primary check.
• ggplot structural .rds snapshots are maintained as fallback fixtures.
• Missing structural fixtures are written only when the visual comparison passed, when structural update mode is enabled, or for newly generated visual snapshots.
• Structural fixtures are not created after a visual mismatch against an existing visual snapshot.
• If the visual check fails but the structural fallback passes, the test warns with the .new.svg review path and asks the user to accept the visual fixture so future runs do not keep using the slower fallback.
• Stale .new.svg files are no longer treated as current mismatches.
• Plot snapshot rendering runs a small opt-out GC step by default to reduce large generated-test slowdown: options(jaspTools.plotSnapshot.gc = FALSE) disables it.

Validation

• testthat::test_dir(tests/testthat, reporter = summary) passes.
• git diff --check passes, apart from existing CRLF warnings on Windows.
• CI-style rcmdcheck::rcmdcheck(args = c(--no-manual, --as-cran, --no-examples), error_on = never) has 0 errors and only the existing allowed warning/notes.
• Verified against the slow jaspMetaAnalysis forest edge-case file; runtime dropped from the pathological fallback behavior to normal visual-check runtime, with an additional GC-related speedup.

@vandenman could you review this plot snapshot fallback behavior?

[FBartos]

Added the companion fix for the eager-decoded result-state contract in jaspBase.

Changes here:

• The fast test plot writer now uses jaspBase's internal decoded-plot materializer when available, so fast replay tests do not persist raw encoded plot objects.
• moveJaspHtmlToDir() now creates the html output directory before copying assets. This avoids the file.copy(...): more 'from' files than 'to' files failure when running runAnalysis(..., view = TRUE) in a fresh Rscript temp directory.
• Added focused tests for both paths.

Verified with:

• test-runAnalysis-fast-test-plots.R
• test-view.R
• A MixedModelsGLMM replay from Larks and Owls.jasp with view = TRUE using a stub viewer; returned state contains no encoded JASP column tokens.

Related jaspBase work: jasp-stats/jaspBase#204.

[FBartos]

@vandenman this is ready for your review. The jaspTools side remains orchestration-only; the paired jaspBase provenance fix has been pushed and the focused bridge/replay checks pass locally.

[vandenman]

This overwrite of jaspBase writeImageJaspResults should be reverted. The other changes to the snapshots make sense. Note that a lot of slowdown in plots has been fixed by updating a dependency (rlang::hash was extremely slow, since fixed on devel).