feat: vcov and precision matrices + move float ops to f64

[Neclow]

Features

• Added variance-covariance matrix and precision matrix calculations for phylo2vec vectors/matrices
• Closes Precision and variance-covariance matrices #87

Refactors

• Move all float types in Rust to f64. This avoids many conversions in R and Python, where base R and numpy use float64 by default, and even Rust too https://github.com/rust-lang/rfcs/blob/master/text/0212-restore-int-fallback.md.
  • This was important to ensure precision of the new functions (precision and variance-covariance functions), where tests become imprecise in f32.
  • However, as mentioned in Make rust module accept f32/f64? #83, a more flexible option which accepts both remains desirable.
• Python: Rename metrics to stats: metrics never made a lot of sense (cophenetic distances are not really a "metric" of a tree, so currently renamed to stats which will contain all sort of operations either between nodes (nodewise in the Python package) or between trees (treewise, upcoming...).
• Update ndarray version in extendr to match the one used in the Rust API

Notes

• Using scipy.linalg.solve(..., assume_a="pos") instead of np.linalg.solve in precision is 50% faster for n = 1000 leaves

[Copilot]

Pull Request Overview

This PR adds variance–covariance (vcv) and precision matrix support to both R and Python bindings, consolidates all floating-point operations in Rust to f64, and renames the “metrics” module to “stats” throughout the codebase.

• Introduce vcv_from_vector, vcv_from_matrix, and precision functions (with Schur complement) in R and Rust, plus corresponding Python cov and precision APIs.
• Convert all Rust matrix and graph code from f32 to f64 to align with downstream R/Python and improve numerical precision.
• Rename the R/Python “metrics” module to “stats”, update imports/tests, and remove obsolete test_metrics.R.

Reviewed Changes

Copilot reviewed 32 out of 33 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
r-phylo2vec/tests/testthat/test_stats.R	Add R tests for vcv and precision functions
r-phylo2vec/src/rust/src/lib.rs	Switch f32 → f64, implement new vcv_from_* and pre_precision_from_* exports
r-phylo2vec/R/stats.R	Add R wrappers for vcv and precision (Schur complement)
py-phylo2vec/phylo2vec/stats/nodewise.py	Rename metrics → stats, implement cov and precision APIs
py-phylo2vec/src/lib.rs	Update Python bindings for f64 arrays and new functions

Comments suppressed due to low confidence (2)

py-phylo2vec/phylo2vec/stats/nodewise.py:83

• The implementation references core.vcv and core.vcv_with_bls but core is never imported; add an import such as import phylo2vec._phylo2vec_core as core (or from phylo2vec import core) at the top of the file to avoid a NameError.

def cov(vector_or_matrix):

py-phylo2vec/phylo2vec/stats/nodewise.py:3

• [nitpick] The warnings import is unused in this module; consider removing it to keep the namespace clean.

import warnings

[Neclow]

R: renamed "cov" to "vcovp" to avoid collisions with base/stats (cov/vcov) and ape (vcv)

[Neclow]

Note: the benchmarks are slightly worse now for matrix operations due to the f32 -> f64 conversion, although still within a reasonable margin in my opinion.