FunctionGemma export: convert MLIR-text rewrites to StableHLO/DAG transforms

[michalharakal]

The FunctionGemma KV/int8 decode graphs are DSL/DAG-authored, but the graph post-processing in
FunctionGemmaExport.kt is done by regex rewrites of the emitted StableHLO text, not graph passes.
Convert each to the SKaiNET way (a StableHLO/DAG transform):

• bf16 weight emission (rewriteGlobalsToBf16) → a DAG/StableHLO dtype transform (leverage DtypeForwardPropagationPass).
• int8 quant (rewriteGlobalsToInt8 injecting dequant as text) → a real graph quantization pass; keep the host per-row quant writer, emit i8→f32 × scale as graph nodes.
• Dynamic KV dims — retire the sentinel-prime 7919 → x?x string relax (most fragile: a magic prime that must not collide with a real dim/SSA id) → proper dynamic-shape tracing.
• GemmaModel.forwardWithPast attnWithPast (hand-wired single-token attention) → an MHA-with-past module forward.
• refsFor positional sub-module resolution → typed HybridTransformerBlock field access.

Context: PR #245. Tracker: SKaiNET-embedded sl2610-function-calling/docs/GEMMA-KV-INT8.md.