Lower QIR subroutine (non-intrinsic) calls beyond the entry point

**Describe the feature you'd like**
Currently the translator structurally lowers only the entry-point function. Calls to user-defined subroutines that aren't QIS intrinsics either error out as unsupported or rely on the producer to have inlined them. Lower QIR subroutine bodies into OpenQASM 3 subroutine constructs — `def` for routines mixing classical and quantum operands, `gate` for purely-quantum helpers — and emit calls to them at the original call sites.

**How would this feature be used? Please describe.**
Producers that emit modular QIR (multiple `define`s with calls between them, rather than a single inlined entry point) currently force users to manually flatten their QIR before handing it to qirtoqasm. A direct translation path would remove that step and produce OpenQASM 3 output that mirrors the input's modular structure.

**Describe alternatives you've considered**
Inlining at the producer side is the current workaround and works for many cases, but it inflates QIR size and makes debugging harder.

**Additional context**
Scope needs to be defined carefully. Pure-quantum subroutines map naturally to `gate`; mixed subroutines map to `def`. Subroutine return values today are captured at the call site as a single SSA id; full lowering would need to model their structure (single value vs. struct, scalar vs. array, etc.) so the emitted signature matches the call sites.