fix: guard QBMM boundary extrapolation with present(pb_in)

[sbryngelson]

Summary

• Post_process calls s_populate_variables_buffers without the optional pb_in/mv_in arguments, but the QBMM guard (qbmm .and. .not. polytropic) could still evaluate to true, causing access to an absent optional argument — undefined behavior in Fortran.
• Add present(pb_in) to all 6 call sites of s_qbmm_extrapolation in m_boundary_common.fpp.
• Caught by gfortran -fcheck=bounds in the macOS debug CI build.

Test plan

• Compiles with gfortran
• ./mfc.sh precheck passes
• CI macOS debug build passes (currently fails on master for QBMM + post_process)

[github-actions]

Claude Code Review

Incremental review from: f3bb4b6
Head SHA: 4ecd249

New findings since last Claude review:

• Silent skip of QBMM boundary extrapolation may produce wrong results (src/common/m_boundary_common.fpp, all 6 hunks): When qbmm .and. (.not. polytropic) is true at runtime but pb_in/mv_in are absent, the new guard silently skips s_qbmm_extrapolation. This prevents a Fortran runtime error, but ghost-cell QBMM state is left uninitialized/stale without any diagnostic. If a call site reaches this code with QBMM non-polytropic physics active but omits the optional arguments, the boundary condition is silently wrong rather than aborting. An @:PROHIBIT(qbmm .and. (.not. polytropic) .and. .not. present(pb_in), "QBMM non-polytropic boundary requires pb_in/mv_in") would catch erroneous call sites rather than silently degrading physics correctness.