Fix Standard-mode parameter checks and latent bugs in the C generator

[aoymt]

This PR collects six small fixes to the Standard-mode C generator found
during a systematic review (cross-checked against the Python port). All
fixes are confirmed against the full base-mode integration suite
(94/94 tests pass before and after; none of the changes affect the
generated definition files).

Fixes

1. TriangularLattice: t'' unused-parameter check inspected the wrong
   variable (e52627d)
   The spin-model branch checked StdI->tp (already checked as t' on
   the previous line) instead of StdI->tpp, so a specified t'' was
   silently ignored.
   ⚠ Behavior change: spin-model inputs that specify t'' were
   silently accepted before and now abort with the standard
   unused-parameter error.

2. Kagome: duplicated, mislabeled check (aca0821)
   StdI->t was checked twice in the spin-model branch, the second time
   under the wrong label "t0" (the real t0 is checked on the next
   line). The redundant mislabeled line is removed; specifying t still
   aborts, now with a single correctly-labeled message.

3. Tetragonal / Honeycomb: re-print of the consumed V' fallback
   (e6d897d)
   V' only serves as the fallback input for V0'/V1'(/V2') via
   StdFace_InputCoulombV, but these two lattices then re-ran
   StdFace_PrintVal_d on it, logging it as if it were an independent
   parameter. stdout logging is the only observable change; no other
   lattice does this.

4. Wannier90: braceless ifs in the cutoff_*Vec parameter loops
   (74bd911)
   The if guarded only the sprintf, so with box[i][j] unset,
   StdFace_PrintVal_d would run with a stale label and a garbage
   default (NaN_i * 0.5). Unreachable in practice because
   StdFace_InitSite fills box first — output is unchanged — but the
   braces make the intended conditional explicit.

5. Wannier90: tUJ/tUJindx initialised to NULL (5645bb4)
   Prevents freeing uninitialised pointers on early-exit paths.

6. Wannier90: error out when _ur.dat is missing for the spin model
   (115e605)
   The superexchange J = 4t²/U needs the on-site U values; a missing
   _ur.dat previously crashed later instead of giving a clear error.

Not included

A seventh finding — the OmegaIm default is computed as
0.01 * (int)StdI->LargeValue (StdFace_main.c), truncating the
fractional part of LargeValue — changes a physics-facing default and
is proposed separately as its own PR (fix/omegaim-default) so it can
be discussed and merged independently.

Notes for reviewers

• The base-mode integration suite on this branch passes 94/94 both
  before and after the fixes; none of the reference files change.
• The pytest (3.10) CI job may fail on this branch for an unrelated,
  pre-existing reason (tomllib is Python 3.11+ stdlib; fixed on the
  Python development line). This PR intentionally touches only src/.

🤖 Generated with Claude Code