Move RICH sensor QE from npsim.py into epic geometry (optical surface EFFICIENCY property)

## Motivation

The quantum efficiency (QE) curves for the DRICH SiPM and PFRICH HRPPD photocathode sensors are currently hardcoded as explicit lambda/efficiency tables in `npsim.py`. These are physical properties of the sensors and belong in the **epic** geometry repository alongside the rest of the detector description.

## Key finding

`optical_materials.xml` in epic already defines `SensorSurface_DRICH` with a **placeholder** flat efficiency:
```xml

```
This surface is already applied as a `SkinSurface` to each pss volume in `DRICH_geo.cpp`. The infrastructure is in place — we just need to replace the placeholder with real data.

## Proposed design

### Part 1 — epic (`compact/optical_materials.xml`)
- Replace the flat `EFFICIENCY = 1.0` on `SensorSurface_DRICH` with real SiPM (S13361-3050NE-08) QE values, converting λ[nm] → E[eV] via `E = 1239.84 / λ`.
- Add a new `SensorSurface_PFRICH` with HRPPD photocathode QE values.
- Reference `SensorSurface_PFRICH` from `compact/pid/pfrich.xml` on the photocathode element.

### Part 2 — npsim stacking action (`OpticalPhotonEfficiencyStackingAction.h`)
- Add a `SensorSurface` property (a surface name string).
- In `initialize()`, if `SensorSurface` is set, look up the named optical surface, read its `EFFICIENCY` from the `G4MaterialPropertiesTable`, convert photon energy back to wavelength, and populate the existing interpolation table (`m_interp_lambda_values` / `m_efficiency`).
- Keep `LambdaValues` + `Efficiency` as explicit fallback for backward compatibility.

Priority in `initialize()`:
1. `SensorSurface` set → read QE from G4 optical surface MPT
2. `LambdaValues` + `Efficiency` set → use explicit tables (current behaviour)
3. `LambdaMin`/`LambdaMax` + `Efficiency` → linspace fallback (legacy hpDIRC path)

### Part 3 — npsim.py
Replace hardcoded lambda/efficiency tables:
```python
# Before (hardcoded in npsim):
{
"name": "OpticalPhotonEfficiencyStackingAction",
"parameter": {
"LambdaValues": [315e-6, 325e-6, ...], # 17 values
"Efficiency": [0.00, 0.04, ...], # 17 values
"LogicalVolume": "DRICH_(gas|aerogel)",
}
},
```
```python
# After (QE lives in epic):
{
"name": "OpticalPhotonEfficiencyStackingAction",
"parameter": {
"SensorSurface": "SensorSurface_DRICH",
"LogicalVolume": "DRICH_(gas|aerogel)",
}
},
```

## Separation of concerns

| Repository | Responsibility |
|---|---|
| **epic** | Sensor QE physics (what wavelengths does this photocathode detect?) |
| **npsim** | Simulation topology (which radiator volumes feed which sensor surface?) |

`npsim.py` retains only the `LogicalVolume` regex — a simulation choice about where to kill photons early. The QE numbers themselves are geometry data.

## Scope

- DRICH (SiPM) and PFRICH (HRPPD) are the primary targets.
- hpDIRC can be migrated in a follow-up once epic defines a `SensorSurface_DIRC` with the barPMT QE.
- The `LambdaValues`/`Efficiency` fallback must be retained until all detectors have migrated.