[garretfick]

Fixes #439.

When collecting scopes, for statements declared their loop variables in the same scope in which the iterable expression had been collected, unlike assignments and comprehensions which allocate a fresh scope first. As a result, a name used in the iterable resolved to the loop variable itself. Inferring the type of the loop variable then required inferring the iterable, which required inferring the loop variable again, and so on until the stack overflowed. This is triggered by code such as:

for idl in idls:
    for idl in idl.idl_files:
        ...

which appears in rules_ros2's ament.bzl and crashed the server whenever that file was analyzed, e.g. after go-to-definition on py_exec_launcher.

The iterable is evaluated before the loop variables are bound, so allocate a new scope for the targets. Names in the iterable now resolve to the enclosing definition (or report "not defined" if there is none).

Additionally, add a re-entrancy guard to infer_source_expr_assign so that any remaining cycle through an assignment source bails out with Unknown instead of recursing indefinitely.