Return the expanded lattice with and without the computed parameters

[DavidSagan]

parse_and_expand_PALS now returns five trees rather than four.

• full_expanded — the lattice the bookkeeper finishes: every dependent parameter computed and present. Each element carries its ReferenceP, FloorP and s_position, the derived members of every parameter family it uses (Kn1L alongside Kn1, voltage alongside gradient, …) and the non-zero defaults of the groups it carries; each branch is capped with a branch_end Placeholder holding its final reference and floor.
• expanded — that same tree pruned back to the author's inputs. A parameter is kept when it was present before run_element_bookkeeper first ran, or when a post-expand_lattice set wrote it. Everything else the bookkeeper produced is dropped, along with any group left empty and the branch_end elements.

expanded is full_expanded with nodes removed rather than an earlier snapshot, so a parameter present in both holds the same value in both, with every set and ABSOLUTE controller applied. Use it to see the inputs rather than their consequences, or to write a lattice back out without the derived values.

API

The tree that took the old expanded name now takes full_expanded: node_link.full_expanded, build_correspondence_map(original, combined, full_expanded, leftover) and get_lattice_parameter_value(full_expanded, leftover, …). expanded takes no part in the correspondence — it is a pruned copy, not a step in the derivation chain, so its nodes are located by path.

struct lattices gained full_expanded between expanded and leftover, so callers binding it by layout need updating. PALSJulia follows in a companion PR.

Tests

New tests/test_expanded_minus_dependent.cpp (9 cases) pins the boundary between the two trees: an authored ReferenceP the bookkeeper completes, one authored member of a multipole family with the rest derived, an authored RFP gradient plus materialized defaults, and an element holding nothing but its kind. The existing suites follow the rename.

Full suite: 1291 assertions in 236 test cases, all passing.

🤖 Generated with Claude Code