Re-enable the target-resolution + sharpen UI once the feature is ready

[prekabreki]

The Target Resolution + Sharpen controls are currently hidden (display:none on the wrapping .adv-grid, fetchforge/index.html card 05) because the feature is not ready to ship. Commit: see PR #38.

The elements remain in the DOM deliberately — startDownload() reads #target-res.value and #sharpen.checked for both the download and local flows, so leaving them present makes every job send the off state (target_res=0, sharpen=false) with no JS changes.

This issue is the reminder to unhide, and the place to decide the bar for doing so.

Blocked by

• Verify HDR/10-bit passthrough through libplacebo on real HDR footage #33 Verify HDR/10-bit passthrough through libplacebo on real HDR footage
• Verify scale_cuda + sharpen path at runtime on Windows NVENC build #34 Verify scale_cuda + sharpen path at runtime on Windows NVENC build
• Skip the scaler when target resolution equals source (same-size resample is wasted work) #35 Skip the scaler when target resolution equals source
• Add FSRCNNX-8 neural upscale option via libplacebo custom_shader_path #37 Add FSRCNNX-8 neural upscale (the thing that makes the feature worth having)
• Raise CAS_STRENGTH from 0.5 toward 0.7-0.75 #39 Raise CAS_STRENGTH toward 0.7-0.75

Acceptance criteria

• All blockers above closed or explicitly waived with a reason.
• display:none removed from the .adv-grid wrapper and the explanatory comment updated or dropped.
• A real end-to-end job with sharpen on and a target resolution set produces visibly better output than passthrough (not just "runs without error").

Files

• fetchforge/index.html — search for Target Resolution + Sharpen are HIDDEN.

Notes

• Trivial change; the work is the judgment call about whether the blockers are genuinely done.

[prekabreki]

Blocker status after the 2026-08-01 session — 3 of 5 closed:

• Verify scale_cuda + sharpen path at runtime on Windows NVENC build #34 Verify scale_cuda + sharpen at runtime — done, verified on Windows/RTX 4080: upscale, upscale+sharpen and same-size passthrough all produce correct-resolution, correct-frame-count, cleanly-decoding output, and cas measurably alters the picture (not a silent no-op). See .memories/scale-cuda-sharpen-runtime-verified.md.
• Skip the scaler when target resolution equals source (same-size resample is wasted work) #35 Skip the scaler at 1:1 — done.
• Raise CAS_STRENGTH from 0.5 toward 0.7-0.75 #39 Raise CAS_STRENGTH — done, now 0.72.
• Verify HDR/10-bit passthrough through libplacebo on real HDR footage #33 Verify HDR/10-bit passthrough through libplacebo — open, needs real HDR footage.
• Add FSRCNNX-8 neural upscale option via libplacebo custom_shader_path #37 FSRCNNX-8 neural upscale — open.

So this stays blocked, and the two remaining blockers both need real source material rather than code. Note the third acceptance criterion here ("visibly better output than passthrough, not just runs without error") also needs real footage, so #33/#40 and this issue want the same session and the same clips.

Heads-up: tools/issue-ready.py currently cannot see the ## Blocked by list above (#63), so this issue has been showing up as ready work. It is not.