Consider sharing Production(s) in schema generation

### Issue

`exipg` generates lots of duplicate `Production` structs. In `staticXmlSchema.c`:
- **445 identical EE productions** `{687865855, INDEX_MAX, {URI_MAX, LN_MAX}}`
- **154× SE→NonTerm 16**, **153× SE→NonTerm 15**, **133× SE→NonTerm 9**
- **186× qname {3,50} (annotation)**, **134× {3,56} (attributeGroup)**

Each Production = 16 bytes. **Waste: ~25-30 KB out of 287 KB (10%)**

### Fix

Define shared constants, reference via pointer:
```c
static CONST Production EE_PROD = {687865855, INDEX_MAX, {URI_MAX, LN_MAX}};
static CONST Production SE_ANNOT_81 = {503316485, 81, {3, 50}};

// Use in GrammarRules
static CONST Production* xmlscm_prod_91_0[] = {&EE_PROD, &SE_ANNOT_81, ...};
```

**Tradeoff:** Pointer indirection vs memory savings

### Impact

- **Memory:** 25-30 KB savings (important for embedded)
- **Files:** Modify `exipg.c` code generator
- **Risk:** Low - verify with test suite after regenerating grammars

**Priority:** Low - optimization, not a bug

Start in exipg.c to see how the generator emits Production values and GrammarRules, then inspect staticXmlSchema.c for the repeated productions described in the issue. Regenerate grammars and run the test suite; done means shared production references reduce the generated data while tests pass and pointer indirection is acceptable.

Written by the indexing model from the issue text.