update Riemannian Conjugate Gradient tunable

[sChoiKr]

RCG update

• Made RCG solver configurable through RCGSolver fields.
• Added tunable coefficient rules :hager_zhang, :polak_ribiere, :fletcher_reeves, :dai_yuan, :hestenes_stiefel, :conjugate_descent, and :steepest.
• Added tunable restart strategies, including :non_descent, :non_sufficient_descent, and :never.
• Added optional Beale restart support via beale_restart.
• Added parameters for restart_threshold, sufficient_descent_kappa, and denom_threshold.
• Updated RCG to pass coefficient and restart_condition explicitly to Manopt’s conjugate_gradient_descent.
• Added RCGSolver metadata to solver_info, including selected coefficient, restart rule, transport type, and thresholds.
• Updated solver dispatch so cpd(...; solver=:rcg, coefficient=..., restart=...) forwards options into RCGSolver.
• Removed the need for local SegreProjectionTransport implementation, relying on upstream Manifolds.jl support.