feat: init binned spectrum and add axis conversions

[jamesrafe]

Adds a BinnedSpectrum type and unit-aware spectral axis conversions.

BinnedSpectrum

A new Spectrum kind for binned data (e.g., X-ray spectra): the spectral axis is an $m \times 2$ matrix of bin edges and the flux axis is an $m$-length vector.

spec = spectrum([1.0 2.0; 2.0 3.0; 3.0 4.0]u"keV", [1.0, 2.0, 3.0]u"Jy")

• Constructed through spectrum() for both unitless and unitful inputs, with validation: exactly two edge columns, and each row strictly ordered in the direction of the spectral axis (catches swapped, mixed-order, and zero-width edges; descending axes are supported).
• Indexing (spec[i] for any Integer, ranges, masks), iteration, and display work like the other spectrum kinds; redshift/doppler_shift handle bin edges.
• redden/deredden reject binned spectra with an explicit error (extinction laws need a bin-integration convention) instead of failing inside DustExtinction.

Axis conversions

Unitful.uconvert is extended for spectra, converting between wavelength, frequency, and photon energy via the photon equivalence $E = hν = h c / λ$ (powered by UnitfulEquivalences.jl):

uconvert(u"THz", spec) # Convert the spectral axis, flux carried unchanged
uconvert((u"THz", u"Jy"), spec) # Also convert flux density
uconvert(unit(other_spec), spec) # Match another spectrum's units

• Same-dimension targets are plain unit conversions, so conversions are idempotent.
• Conversions are elementwise: a reciprocal conversion (e.g. wavelength → frequency) yields an axis with the opposite direction of monotonicity.
• Flux density conversion uses the new SpectralDensity(at) equivalence (re-exported here), which preserves integrals: $F_ν = F_λ λ² / c$. It is undefined for binned spectra (bin values are integrated quantities) and errors accordingly.
• redshift/doppler_shift now also handle frequency and energy axes (compressing instead of stretching).

Dependencies

We are seeing if SpectralDensity can be upstreamed to sostock/UnitfulEquivalences.jl#31.

[fjebaker]

[codecov]

Codecov Report

❌ Patch coverage is 97.87234% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 96.80%. Comparing base (e3b22ad) to head (862f9da).

Files with missing lines	Patch %	Lines
src/transforms/spectral_axis.jl	92.85%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #61      +/-   ##
==========================================
+ Coverage   95.00%   96.80%   +1.80%     
==========================================
  Files           9       11       +2     
  Lines         180      219      +39     
==========================================
+ Hits          171      212      +41     
+ Misses          9        7       -2     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[icweaver]

This is a really nice PR, thanks @jamesrafe! I think this would be a good opportunity to flex our ecosystem here and start factoring out some common functionality I've been seeing pop-up in a few of our packages.

For starters: @cgarling, what would you think about upstreaming your spectral density conversions in PhotometricFilters.jl to UnitfulEquivalences.jl? I've opened a PR to demonstrate what that might look like sostock/UnitfulEquivalences.jl#31

Next, I think we could replace the custom to_wavelength/to_frequency API introduced in this PR with simple methods extending Untiful.jl's uconvert. Something like this seems ergonomic to me:

uconvert(u"THz", spec) # Just convert the spectral axis
uconvert((u"keV", u"W/m^2/keV"), spec) # Convert the spectral AND flux axis

compared to the separate with_spectral_axis_and_flux_units()/with_spectral_axis_unit()/with_flux_unit() that astropy/specutils seems to juggle

There's also some cleanup in the binned spectrum logic I think we could apply here. This will all be included in my review, but just wanted to give y'all a heads up first on where I was thinking of taking this

[cgarling]

No problem with moving the spectral density conversions if they accept the pr, IMO its a little more specialized than their other equivalences so IDK if they will want to add it

[icweaver]

I may have been spoiled with the DynamicQuantities dev being so receptive to some of our astro-specific PRs in the past, so will see I guess 😅

[icweaver]

Ok, I think the bones are in place now. Instead of a separate review at this stage, I've just added a PR description to summarize where things are at. It depends on one moving target, the potential PR to UnitfulEquivalences.jl for our unit conversions if they are amenable, but I think the other main piece of this PR, the addition of BinnedSpectrum, is ready for another pass. I just filled in (862f9da) a few of the remaining gaps in the implementation, which are summarized in the above PR description along with James' other really nice additions.

Please have a look all, and feel free to make any additional suggestions if you have any at this time. I know I personally want to polish up the plotting bits, but I think that can be saved for a future PR.