Add search site and printable QR location stickers

[sr320]

Phase 2: make the inventory findable before anyone is asked to enter data into it. Search value has to land first or this goes the way of the spreadsheet.

build_site.py inlines the data into a single self-contained _site/index.html rather than fetching a JSON sidecar, so the built page works over file:// -- no local server to preview it, and no CORS surprises.

No search library. At 499 items, a precomputed lowercase haystack per item plus a scored linear scan is faster than any index would be, and there is nothing vendored to keep alive. Every query token must match somewhere; name-prefix scores above name-substring above any-field. Location labels are part of the haystack, so "fridge door" and "cabinet 3" are queries too.

Location deep links (#209-CAB-01) roll up descendants: scanning the sticker on the -20 freezer shows all 49 items across its shelves and drawers, not the zero items filed directly against the freezer itself. That rollup is the whole point of the QR path.

make_qr_labels.py emits HTML, not PDF, so there is no PDF toolchain to maintain -- open it and print. Sticker geometry lives in CSS variables at the top of the generated file so it can be nudged to fit label stock without touching the script. QR payloads in CI take their base URL from configure-pages output rather than a hardcoded string.

Two bugs found by actually loading the page instead of assuming:

• The data placeholder was /*__INVENTORY_DATA__*/null but only the comment was substituted, leaving const DATA = {...}null; -- a syntax error that silently produced a blank page. The placeholder now swallows the null, and the template stays valid JavaScript.
• segno's default SVG has width/height but no viewBox, so the CSS sticker size only enlarged the viewport while the code stayed at intrinsic size in the corner. omitsize=True emits a viewBox instead.

Also tightened three legacy-migration warts that were visible in the UI: multiple parenthetical dates now keep the earliest as received and preserve the rest in notes; unbalanced parentheses left by date stripping (and present in the source, e.g. "2-Mercaptoethanol (8/6/2018) (opened (8/8/2018)") move their fragment to notes instead of shipping a name ending mid-parenthesis; and documentation now outranks content words, so "DNeasy blood and tissue handbook" is office rather than a blood sample, while "sample tubes" is a container rather than a sample.

Pages deploy needs one manual step this workflow cannot do for itself: Settings > Pages > Source = GitHub Actions.