Renormalizer interface for TRG (attempt 2)

[Yue-Zhengyuan]

Motivation

• The current run! interface mixes algorithm configuration (truncation, stopping criterion, finalizers) with execution.
• Intermediate states are inaccessible without writing custom Finalizer callbacks. This makes it hard to inspect what happens during an RG flow or to capture observables mid-run.

What this PR adds

Renormalizer — an iterator that separates algorithm parameters from tensor state:

params = TRGParams(; trunc = truncrank(24), stop = maxiter(25))
renorm = Renormalizer(params, T)

# Iterate: each step yields (state, data)
for (state, data) in renorm
    τ, _ = extract_tau_and_c(state.T; fast = false)
    # inspect observables or break at any step. no need for finalizers.
end

# Or advance one step at a time
rgstep!(renorm)

# Or run to completion
state, data = run!(renorm; verbosity = 1)

Architecture

TNRParams        ← pure algorithm config (truncation, stopping criterion)
Renormalizer     ← RG iterator (params + state + data + step counter)
TRG{E,S,TT}      ← unchanged existing type, now serves as Renormalizer state
TRGParams        ← TRG-specific params (@kwdef, sensible defaults)

Backward compatibility

The 3-argument run! for TRG is preserved as a thin wrapper that constructs TRGParams + Renormalizer internally and delegates:

scheme = TRG(T)
data = run!(scheme, truncrank(16), maxiter(20))  # unchanged

The new interface is a parallel layer that can be adopted incrementally.

• Finalizer support is removed from TRG's run!. Also, the initial tensors are always normalized (finalized). Users who previously used a Finalizer to capture per-step data can instead use the iterator interface, which gives direct access to intermediate states without callback machinery. The Finalizer infrastructure remains in the codebase for other schemes (LoopTNR, etc.) and is tagged with TODO markers for eventual migration.

Side improvements

• gs_degeneracy.jl: tensor-level methods + shared helper (huge deduplication)
• cft.jl: extracted _row_transfer_matrix helper, reused by CFT and GSD paths
• Tests: free energy, intermediate τ and CFT data captured in a single RG run instead of running twice

[Yue-Zhengyuan]

@VictorVanthilt I have made substantial efforts to maintain backward compatibility. This PR should be much more acceptable.

[codecov]

Codecov Report

❌ Patch coverage is 41.66667% with 49 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/utility/gs_degeneracy.jl	0.00%	28 Missing ⚠️
src/schemes/renormalizer.jl	54.83%	14 Missing ⚠️
src/schemes/trg.jl	63.15%	7 Missing ⚠️

Files with missing lines	Coverage Δ
src/schemes/atrg.jl	73.33% <ø> (ø)
src/schemes/atrg3d.jl	0.00% <ø> (-95.56%)	⬇️
src/schemes/btrg.jl	84.61% <ø> (ø)
src/schemes/ctm/c4vctm.jl	90.27% <ø> (ø)
src/schemes/ctm/honeycomb.jl	0.00% <ø> (-93.34%)	⬇️
src/schemes/ctm/triangular.jl	0.00% <ø> (-67.40%)	⬇️
src/schemes/hotrg.jl	96.96% <ø> (ø)
src/schemes/hotrg3d.jl	0.00% <ø> (-93.66%)	⬇️
src/schemes/impurityhotrg.jl	0.00% <ø> (-100.00%)	⬇️
src/schemes/impuritytrg.jl	0.00% <ø> (-100.00%)	⬇️
... and 9 more

... and 23 files with indirect coverage changes

🚀 New features to boost your workflow:

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