perf: speed up codegen (#48)

---------

Co-authored-by: Adam Chidlow <achidlow@users.noreply.github.com>

src/puya/codegen/context.py

@@ -1,10 +1,31 @@
 import attrs
 
+from puya.codegen import ops as mir
+from puya.codegen.vla import VariableLifetimeAnalysis
 from puya.context import CompileContext
 from puya.ir import models as ir
+from puya.utils import attrs_extend
 
 
-@attrs.frozen
+@attrs.define
 class ProgramCodeGenContext(CompileContext):
     contract: ir.Contract
     program: ir.Program
+
+    def for_subroutine(self, subroutine: mir.MemorySubroutine) -> "SubroutineCodeGenContext":
+        return attrs_extend(SubroutineCodeGenContext, self, subroutine=subroutine)
+
+
+@attrs.define(frozen=False)
+class SubroutineCodeGenContext(ProgramCodeGenContext):
+    subroutine: mir.MemorySubroutine
+    _vla: VariableLifetimeAnalysis | None = attrs.field(default=None)
+
+    @property
+    def vla(self) -> VariableLifetimeAnalysis:
+        if self._vla is None:
+            self._vla = VariableLifetimeAnalysis.analyze(self.subroutine)
+        return self._vla
+
+    def invalidate_vla(self) -> None:
+        self._vla = None

src/puya/codegen/ops.py

@@ -227,6 +227,7 @@ def __str__(self) -> str:
         return f"pop {self.n}"
 
 
+@t.final
 @attrs.frozen(eq=False, init=False)
 class VirtualStackOp(BaseOp):
     original: Sequence[BaseOp]