fix(site): stop showing 'could not load' during normal chunk streaming

[datascry]

on mobile it says could not load roles database for about 5 seconds until the first json loads.. that is misleading

You were right. The results-status template branched on three of four `DbStatus` values:

{#if dbStatus === "ready"}        ...
{:else if dbStatus === "loading"} LOADING ROLES…
{:else}                           COULD NOT LOAD THE JOB DATABASE.
{/if}

`dbStatus = "loading-progressive"` — the legitimate state where chunk 0 has merged and chunks 1-N are streaming in — fell through to the `{:else}` branch and surfaced the alarming "COULD NOT LOAD" copy. On mobile that branch lit up for ~5 s during every cold load while chunks streamed, with actual results visible underneath. Misleading.

Fix

Invert the branching: error gets its own explicit case, the catch-all shows the loading indicator. Same fix on the `aria-busy` attribute so assistive tech reflects the in-progress state during chunk streaming.

State	Before	After
"loading" (waiting on manifest + chunk 0)	LOADING ROLES… ✓	LOADING ROLES… ✓
"loading-progressive" (chunks streaming)	COULD NOT LOAD ✗	LOADING ROLES… ✓
"error" (real terminal failure)	COULD NOT LOAD ✓ (but ambiguous with above)	COULD NOT LOAD ✓ (now unambiguous)
"ready"	"N ROLES · PAGE 1" ✓	"N ROLES · PAGE 1" ✓

Test plan

• bun run typecheck clean
• Built bundle inspected — branching is now ready / error / else with else covering both loading states
• CI green
• Post-deploy mobile check: open openroles.today on a phone, watch the results-status line during the first 5 seconds of cold load — should read "LOADING ROLES…" with the busy dot, never "COULD NOT LOAD"