engine: MDL writer prints wildcard subscripts as '*' instead of 'Dim!' -- exported .mdl fails re-parse #847

Problem

The MDL writer emits XMILE wildcard syntax inside subscript lists, producing .mdl files that simlin's own MDL parser rejects.

Vensim SUM(a[DimA!]) imports as sum(a[*]) (the bang-subscript reduction is canonicalized to an XMILE wildcard). On export, MdlPrintVisitor::walk_index in src/simlin-engine/src/mdl/writer.rs (~line 627) prints IndexExpr0::Wildcard as *:

fn walk_index(&mut self, expr: &IndexExpr0) -> String {
    match expr {
        IndexExpr0::Wildcard(_) => "*".to_string(),
        ...

But MDL subscript lists require the dimension-bang form (DimA!). Re-parsing the exported file fails with:

expected symbol in subscript list, found *

Root cause

The expression printer is context-free: it has no dimension table, so it cannot reconstruct which dimension the wildcard ranges over. Fixing this requires threading per-variable dimension context into the writer so a[*] on a variable arrayed over DimA prints as a[DimA!]. (Multi-dimensional cases need positional mapping of each wildcard slot to the corresponding dimension.)

Impact

A corpus sweep (parse -> project_to_mdl -> re-parse over all test/**/*.mdl under 200KB) shows ~17 fixtures fail re-parse from this alone, out of 29/240 total re-parse failures (the remainder includes the zero-arg builtin class tracked separately). Affected fixtures include:

• test/sdeverywhere/models/sum/sum.mdl
• test/sdeverywhere/models/vector/vector.mdl
• test/sdeverywhere/models/except/except.mdl
• test/test-models/tests/subscript_aggregation
• test/metasd/scientific-revolution/scirev7.mdl

Severity: critical for MDL export of any arrayed model using bang-subscript reductions (SUM/PROD/VMIN/VMAX/etc.) -- the exported file is unusable in Vensim and unreadable by simlin itself.

Component

src/simlin-engine/src/mdl/writer.rs (MdlPrintVisitor::walk_index, project_to_mdl equation path)

Possible approach

Pass the variable's resolved dimensions into the print visitor (similar context is already available where equations are written per-variable), and map each wildcard subscript position to format_mdl_ident(dim_name) + "!". A corpus-wide parse -> export -> re-parse sweep would make a good regression gate.

Related: the second re-parse failure class (zero-arg builtins like PI()) is filed separately.

Context: identified during an MDL writer round-trip corpus sweep.