Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@crates/codegen/src/compile.rs`:
- Around line 5460-5477: The list→tuple optimization should be skipped when
compiling the iterable of an async for so GET_AITER sees the original list type;
modify the block that currently checks ast::Expr::List (the one that compiles
elts and emits Instruction::BuildTuple) to also verify we are NOT on the
async-for path (e.g., add a guard like !self.is_compiling_async_for() or check
the parent node kind) before building the tuple; if in an async for, fall
through to the existing self.compile_expression(iter)? branch so the list
literal is preserved and error messages remain source-faithful.

In `@crates/codegen/src/symboltable.rs`:
- Around line 300-301: The branch checking symbol_table.needs_classdict and
symbol_table.future_annotations can be wrong for nested class symbol tables
because SymbolTable::new() defaults future_annotations to false and finish()
only sets the top-level module; propagate the parent's future_annotations into
any nested SymbolTable instances when creating them (or have finish() propagate
the flag into child class tables) so nested class tables correctly see the from
__future__ import annotations flag before this check; update the code that
constructs nested symbol tables (where SymbolTable::new() is called for classes)
to copy parent.symbol_table.future_annotations into the child
SymbolTable.future_annotations, ensuring needs_classdict logic uses the correct
flag.

---

Nitpick comments:
In `@crates/codegen/src/compile.rs`:
- Around line 1799-1806: Extract the duplicated module-level
__conditional_annotations__ setup into a single helper (e.g., add a method like
emit_module_conditional_annotations or init_conditional_annotations) and call it
from both compile_program and enter_scope; specifically, move the logic that
checks has_module_cond_ann, inserts "__conditional_annotations__" into
current_code_info().metadata.cellvars, and emits the MAKE_CELL sequence into
that helper so both code paths (the open‑coded path in compile_program and the
one in enter_scope) invoke the same routine, passing only the differing value
(has_module_cond_ann or the current scope) as needed to avoid duplication.

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@crates/codegen/src/compile.rs`:
- Around line 1113-1116: The module conditional-annotation preamble must be
emitted identically for both entry paths; modify compile_program_single() to
share the same logic used by compile_program() by checking
ste.has_conditional_annotations (or computing the conditional-annotations flag
first) and then invoking the common preamble emissions that insert the
root-module cell into cellvar_cache (insert "__conditional_annotations__") and
emit the MAKE_CELL preamble instead of falling through to RESUME; refactor so
the differing entry value is determined first and the shared logic that updates
cellvar_cache and emits the MAKE_CELL preamble is called from both
compile_program() and compile_program_single(), ensuring bytecode parity (also
apply same refactor to the other occurrence around the code referenced at
~1799-1817).