fix(instantiate): preserve nested array modifier element scope

[hechuan9]

Summary

• Preserve the current array element's scope when distributing nested, non-each modifiers during component-array expansion.
• For Pump pumps[2](per(curve(eta=motorEta))), the two scalar instances now retain motorEta[1] and motorEta[2] respectively instead of both retaining the unsliced matrix.
• This addresses only the array-modifier element-scope subproblem tracked in ClimaMind CLI-37. It does not include end lowering or downstream FMI compensation.

Spec / MLS Alignment

• Relevant active spec(s) checked: SPEC_0007, SPEC_0021, SPEC_0022, SPEC_0025, SPEC_0029, SPEC_0032.
• Relevant MLS section(s), if semantics changed: MLS §7.2 (modifier value context) and §7.2.5 (each and array-modifier distribution).
• Crate/phase owner: rumoca-phase-instantiate, array expansion.

Risk and Design Notes

• Main correctness risk: indexing scalar or shape-unproven nested expressions. The implementation reuses the existing conservative index_array_expression_for_element path and leaves values unchanged when array projection cannot be proven.
• Main maintenance risk: one additional private recursive helper in the existing array-expansion owner.
• Why the change belongs in these crate(s): the first divergence occurs while an array component is expanded into scalar instances; later Flat/DAE/FMI layers should receive already-correct modifier bindings.
• Any new abstraction, public API, or migration path: one private helper; no public API or migration.

Testing

• Focused regression: cargo test -p rumoca --test mod_propagation_test test_nested_modifier_on_array_component_selects_element_row -- --exact
• The regression proves elements 1 and 2 select rows 1 and 2, while nested each preserves the complete shared array for both elements.
• CI on exact head 0d9d172 passes formatting, lint, documentation, coverage, Linux/macOS/Windows tests, all four MSL quality shards and the final MSL merge gate.

Code Size Budget (required)

• production_lines_added: 45
• production_lines_deleted: 0
• test_lines_added: 65
• test_lines_deleted: 0
• public_items_added: 0
• public_items_removed: 0
• files_touched: 2
• net_added_lines: 110

The net growth is the private recursive traversal needed to reach nested modification leaves plus one end-to-end regression. The first compression pass reduced the prior downstream candidate from 249 added lines across three files to 110 lines across two files by removing unrelated modifier-environment behavior and consolidating recursion around the existing conservative array-projection helper. No follow-up cleanup is planned because no parallel path or public abstraction remains.

Reviewer Checklist

• Relevant active specs were checked.
• MLS-sensitive changes cite the right MLS section.
• Crate boundaries and phase ownership preserved (SPEC_0029).
• Tests prove behavior or explain the remaining gap.
• Standard CI gates pass (fmt, clippy -D warnings, cargo test, cargo doc).
• MSL gate run for compiler/simulator changes; no regression vs baseline.
• Size-budget section completed.
• Positive net diff has explicit compression justification.
• New APIs are required and minimal.
• Old/new parallel paths removed unless explicitly migrating.
• No #[allow(clippy::...)] added outside generated code.
• Every commit signed off (git commit -s); no Co-Authored-By for AI.
• External material (if any) attributed and Apache-2.0 compatible.