Make extending the schematic by clicking pins discoverable and reliable

[xtofalex]

Extending the view from an instance's pins was hard to discover and to use:

• no hover feedback, and on a partial-interface box every visible pin's net was already shown, so double-clicking one did nothing useful;
• the hit test took the first pin within 18px instead of the nearest, so zoomed out a click often landed on the neighbouring pin;
• re-requesting a shown net (or a trace re-listing it) added it twice, and the duplicate wires were then drawn as a fake bus;
• with hierarchy frames on, each added net re-laid out the drawing and the box the user was working on jumped away.

Changes:

• Pins whose net isn't in the view are "open" (hollow-circle stub). A single click adds their net, and the stub fills while it loads. A click on a merged bus pin shows its bits. Box-body double-click still reveals the full interface, and after a short rest a tooltip says so.
• Hovering a pin highlights it, shows a hand cursor when clicking does something, and shows a tooltip naming the action.
• Nearest-pin hit testing (SchematicInteraction::pickPin), with the radius capped in world units so pins don't swallow box-body clicks at low zoom.
• In-flight pin requests are tracked (PendingRequests, 5 s expiry), and only the first press of a double-click acts.
• When the net arrives, the view pans so the clicked box stays put on screen.
• GUIData::addEquipotential() skips nets whose endpoints are all shown, and identical wires are only drawn as a bus between real bus pins.

Tests: SchematicInteractionTest (hit testing, pending requests), GUIDataTest (net dedupe), and EquipotentialViewTest, which drives the real renderSchematic() in a headless ImGui context with simulated mouse input.