explorer: point-mode counts use padded fetch box, not visible viewport

[rdhyee]

Follow-up suggested by Codex on PR #210 (which closes #206 by surfacing the real point-mode count when capped).

Current state after #210

Point-mode "Samples in View" displays the real count from a count(*) query against the padded fetch box (30% larger than the visible viewport, explorer.qmd:1517-1522), not the actually-visible bounds. The cap is no longer silent — users see e.g. "23,422 samples in view (showing 5,000 — zoom in for more)" — but "in view" still loosely means "in the padded cache window" rather than "in the visible Cesium camera frustum."

The same loose meaning was true before #210; the fix didn't tighten it, just stopped the under-reporting.

Why this matters

A small pan inside the cached padded window currently:

• Renders from cache (no re-fetch — cheap),
• Re-displays the cached total count (post-explorer: report real Samples-in-View count when cap is reached (closes #206) #210 Codex fix — stable),
• But that total is for the original padded fetch area, not for what the user can see now.

In dense regions where the padded box is much larger than the visible viewport (e.g., Cyprus at alt=15212), this overstates the visible count by whatever fraction of the padded box is off-screen.

Proposed fix

When point-mode samples are loaded:

• Keep the padded fetch + cache (good for smooth panning).
• But compute "Samples in View" against the visible bounds, not padded.
• On cache-hit pans, recompute the visible-bounds intersection cheaply on the cached array (no DuckDB round-trip required for the rendered count; the total can be re-queried or estimated).

Possible structure:

• cachedTotalCount (post-explorer: report real Samples-in-View count when cap is reached (closes #206) #210) becomes "total samples in padded fetch box, capped at POINT_BUDGET-renders" — internal.
• New inViewCount(bounds) filters cachedData by visible bounds for the rendered tally.
• When capped, the displayed "Samples in View" should be the visible-bounds count, not the padded-box count. May require a small additional count query keyed on the visible bounds.

Acceptance

• "Samples in View" displayed value matches what the camera actually shows (within rounding).
• Cache-hit pans update the stat boxes to reflect the new visible bounds without re-fetching.
• No regression on capped/uncapped distinction from explorer: report real Samples-in-View count when cap is reached (closes #206) #210.

Refs

• explorer: 'Samples in View' counter is the fetch budget, not the real count (#201 Part 1) #206 / explorer: report real Samples-in-View count when cap is reached (closes #206) #210 — original counter accuracy work
• explorer: fix URL state round-trip (closes #201 parts A+B) #203 retrospective Codex review — surfaced this distinction
• explorer: refactor URL/mode state management (centralize writers, collapse dual mode) #208 — the broader state management refactor; this issue is a candidate to fold in there if scope grows