Fix synthetic filter interpolation outside passbands

[nialljmiller]

This fixes synthetic photometry for filters interpolated onto SED wavelength grids.

Previously, interpolated filter transmission could remain nonzero outside the tabulated filter wavelength range.
This produced out-of-band flux leakage for hot/blue SEDs, especially WD and blackbody models, causing unphysical synthetic colors. I found this testing WD cooling tracks.

The fix clamps interpolated filter transmission to zero outside the original filter support and clips negative interpolated transmission values to zero.

Here is the function that has been added to colors/private/synthetic.f90 --

      fmin = minval(filt_wave)
      fmax = maxval(filt_wave)

      do i = 1, size(target_wave)
         if (target_wave(i) < fmin .or. target_wave(i) > fmax) then
            filt_on_target(i) = 0.0_dp
         else if (filt_on_target(i) < 0.0_dp) then
            filt_on_target(i) = 0.0_dp
         end if
      end do

[Debraheem]

Nice find! I can't look at my computer for another couple days because im moving apts, just an fyi! But this is awesome!

Is there a reason the bolometric flux tests didn't capture this in the unit tests, maybe because i used too low log G and low teff? Do we need a high logG hot Teff bb flux unit test added to the others, so this would be caught?

Is there a particular place where this comes? Perhaps we add a minor changelog entry for this to mark it and tweak the unit tests slightly to capture this scenario?

[nialljmiller]

I have made this a draft while I verify the fix works across a few grids. Currently, the koester grid is giving a large offset at unpredictable points.

[nialljmiller]

Summary

This PR fixes numerical issues in the Colors synthetic-photometry machinery exposed by white-dwarf cooling tracks and hot/blue SED tests.

The main bug was that interpolated filter transmission curves could remain nonzero outside the tabulated filter passband. This allowed out-of-band flux to leak into synthetic magnitudes, especially for hot SEDs with substantial blue/UV flux. The fix enforces compact filter support after interpolation: transmission is set to zero outside the original filter wavelength range, and negative interpolated transmission values are clipped to zero. This is applied both to the model-star convolution and to the Vega zero-point calculation.

The PR also makes the SED interpolation machinery more robust. It adds a bounded Hermite interpolation path, fixes Hermite behavior for singleton grid axes, and simplifies the linear interpolation path so degenerate axes and grid-edge cases are handled consistently.

Main changes

• Added zero_filter_outside_support in colors/private/synthetic.f90.
  This prevents both model-SED and Vega/reference-SED flux leakage outside the tabulated filter passband.

• Added colors/private/hermite_interp_bounded.f90.
  This provides construct_sed_hermite_bounded, supports both preloaded flux cubes and individual SED-file fallback, vectorizes interpolation over wavelength, and uses a local multilinear SED as a safety reference.

• Added bounded-Hermite safeguards.
  Hermite remains the preferred result, but invalid, nonpositive, NaN, or excessive local overshoot values fall back to multilinear interpolation. A whole-SED flux-scale check also falls back to multilinear interpolation when the Hermite SED differs too strongly from the multilinear reference.

• Fixed singleton-axis handling in both hermite_interp.f90 and hermite_interp_bounded.f90.
  Axes with one grid point are now treated as valid degenerate dimensions, with zero interpolation extent and zero derivative contribution, rather than as out-of-grid failures. This is important for grids such as Koester white-dwarf atmospheres with a single metallicity axis.

• Made bolometric.f90 default to Hermite_bounded.
  The default fallback also now uses bounded Hermite. calculate_bolometric_phot now cleans a local flux copy before integration instead of mutating the input SED arrays.

• Simplified linear_interp.f90.
  The old explicit 8-corner/log-space path was replaced with a shared setup_axis helper and a dimension-agnostic weighted loop. This handles singleton axes, edge clamping, and interpolation weights consistently.

• Updates colors/Makefile to compile hermite_interp_bounded.f90.

• Updated star/test_suite/custom_colors/python_helpers/test_timings.py.
  Removed the obsolete Newton-step Colors configurations and adjusted the timing benchmark/feature-breakdown plots to the remaining supported configurations.

• Updates colors/README.rst to include comments on interpolation issues.