Unused locals in m_riemann_solver_lf.fpp, one occupying a GPU private() slot

[sbryngelson]

Summary

src/simulation/m_riemann_solver_lf.fpp declares two locals that are never read. One of them also occupies a slot in a GPU private() clause, so every device thread carries storage for a value nothing uses.

Detail

variable	occurrences in the file	what they are
H_avg	1	the declaration only
c_sum_Yi_Phi	2	the declaration, plus one private() clause entry

real(wp) :: Cp_avg, Cv_avg, T_avg, eps, c_sum_Yi_Phi          ! line 62
real(wp) :: H_avg                                             ! line 76
...
$:GPU_PARALLEL_LOOP(collapse=3, private='[..., eps, c_sum_Yi_Phi, Cp_L, Cp_R, ...]')

Neither is ever assigned or read.

Not a regression

Occurrence counts are identical on master, so both predate the current work. Found while reviewing #1714 (the eos_state refactor), which touches this file but did not introduce them — likely leftovers from when the Lax-Friedrichs solver was factored out of a solver that did compute a Roe-averaged enthalpy and a chemistry sound-speed term, both of which the LF path does not need.

For comparison, the sibling solvers do use these: m_riemann_solver_hll.fpp and m_riemann_solver_hllc.fpp both read H_avg and c_sum_Yi_Phi in their Roe-average paths.

Impact

Minor. No functional effect — an unused local is free on CPU. The private() entry is the part with any cost: it reserves per-thread storage on device for a value that is never touched.

Suggested fix

Delete both declarations and remove c_sum_Yi_Phi from the private() clause. Confirm nothing else in the file references them first (a whole-file grep is sufficient given the counts above), and check that no #:if variant reintroduces a use.

Worth checking the other Riemann solvers for the same pattern while in there.