Improve web scanner startup and testing-channel UX

[ShiyunXu]

Summary

• start Catalog v2 loading immediately instead of after both ONNX sessions compile
• download detector and embedder model weights concurrently
• keep multipart loading byte progress monotonic
• support and visibly identify ?channel=testing in the scanner playground and screen capture monitor
• preserve the selected channel when navigating between web tools
• expose the main scanner Settings button clearly and use consistent Minimum corner sharpness labels

The two ONNX sessions are still compiled sequentially to avoid increasing peak memory pressure on mobile browsers.

Why startup appeared hung

There are no artificial startup delays. A cold MTG Catalog v2 load currently downloads about 35 MB compressed (26.0 MB embeddings plus 8.9 MB records), verifies checksums, decompresses and parses 111k records, applies deltas, and writes an IndexedDB snapshot.

Validation

• JavaScript syntax checks for scanner, playground, monitor, and worker
• Catalog v2 tests: 65 passed

[HanClinto]

This is a good point -- there is no reason why it should load all of these long operations sequentially -- it should absolutely be parallelizing them.

[ShiyunXu]

Reopening this, with an apology and an explanation.

This PR — along with a batch of issues and PRs filed here on Aug 25 — was opened by an automated agent benchmark I was running, not by me directly. The harness handed each sandboxed agent a GitHub token that unintentionally carried write scope, so the agents "completed" their tasks by filing real issues and PRs. That was my mistake in configuring the run, and I'm sorry for the noise.

I closed the whole batch a few minutes ago as cleanup. I then saw that you'd already replied here agreeing that the sequential loading should be parallelized, so closing this one silently was the wrong call — hence reopening it and leaving it entirely to you.

To be clear about what this PR is and isn't: the code is machine-generated (+386/-68 across 14 files) and has not been reviewed by me, so please don't treat it as a vetted contribution. If the direction is useful, feel free to take it, cherry-pick from it, or use it as a starting point. If you'd rather not deal with it, just close it and I won't reopen it again.

I've closed the rest of the batch and fixed the token scope so this can't recur.

[HanClinto]

@ShiyunXu Thank you so much for the explanation -- no worries at all! :) That's cool to read about. Thanks for the useful critique, for cleaning up the noise, and I hope my repo has been a useful test bench for you! Best of luck in the rest of your development!