Replace covariance stages with two unified TJPCov v0.5 wrappers

[Copilot]

Five covariance classes with duplicated, hand-rolled Gaussian/NaMaster logic are replaced by two thin wrapper stages that delegate entirely to TJPCov v0.5's CovarianceCalculator. The covariance type and all options move into the pipeline config.

Removed

• TXFourierGaussianCovariance, TXRealGaussianCovariance — hand-rolled Knox formula
• TXFourierTJPCovariance — partial TJPCov integration with TXPipe-side coupling matrix management
• TXFourierNamasterCovariance, TXRealNamasterCovariance — direct NaMaster calls

New classes (txpipe/covariance.py)

• TXFourierCovariance — reads cosmology, tracer metadata, and twopoint_data_fourier; builds TJPCov config; calls CovarianceCalculator; writes summary_statistics_fourier
• TXRealCovariance — same pattern for real space

covariance_nmt.py is emptied with a note; __init__.py and all example configs updated.

Configuration

All TJPCov options are expressed in the TXPipe config. Covariance type is freely composable:

TXFourierCovariance:
    cov_type: [FourierGaussianFsky, FourierSSCHaloModel]
    galaxy_bias: [1.4, 1.5, 1.7]   # per lens bin; default 1.0
    IA: 0.5
    fsky: -1.0                      # auto-computed from tracer_metadata if negative
    outdir: ./cache

TXRealCovariance:
    cov_type: [RealGaussianFsky]

Tracer noise (sigma_e, Ngal) is populated automatically from tracer_metadata. The use_true_shear flag zeroes shape noise for noiseless simulations.

[carlosggarcia]

@joezuntz Following up from the parallel. I think this is the PR you referred to. I had a quick look and I think what you wanted to do (TJPCov doing the heavy lifting) is doable with TXFourierTJPCovariance (

TXPipe/txpipe/covariance.py

Line 609 in f8e86c6

class TXFourierTJPCovariance(PipelineStage):

) and TXRealGaussianCovariance (

TXPipe/txpipe/covariance.py

Line 537 in f8e86c6

class TXRealGaussianCovariance(TXFourierGaussianCovariance):

).

If we see some of these are missing something, I'd build on these, delete all the other options (including the covariance_nmt.py file) and completely ignore copilot in this one.