Indexed assignment incorrectly narrows `Any` values

[dillydill123]

Describe the Bug

Indexing into Any narrows a type:

from typing import Any, reveal_type
values: list[float] = []
frame: Any = object()
reveal_type(frame)
frame["start"] = values
reveal_type(frame["start"])
frame["end"] = frame["start"].shift(-1)  # ERROR: Object of class 'list' has no attribute `shift` [missing-attribute]

This code is obviously bad and won't run, but it shows the error. A value typed as Any should retain Any behavior through indexed access.

I observed this error with replace-imports-with-any, in particular with geopandas. Essentially the real world code was something like:

distance: ndarray[...] = ...
frame = geopandas.GeoDataFrame()
frame["column_a"] = distance
frame["column_b"] = frame["column_a"].shift(-1)  # ERROR: [missing-attribute]

I have geopandas.* in replace-imports-with-any. I think pyrefly sees that frame["column_a"] got assigned an ndarray, and then later thinks frame["column_a"] produces the same ndarray, when geopandas converts it to a Series, of which shift is valid.

Mypy reveals both frame and frame["start"] as Any and reports no issues.

Sandbox Link

https://pyrefly.org/sandbox/?project=v2.lZQxD4IwEIX_ygUGcIDGld0ZwwokIBRoopS0h_Hne23FgBoSJ5Lj3qXt--79G5mbsLFoLEvWURphSYfM6ZadIisTI6GC49R4t1KHtoVq9psXHrGosPDMBDf3R_uqbaXkY-t0nz2xHkSHYXQ8APhwyrI0SyB16yQ7cHEQmNMHMJDvo4QaUYnLjBwqK64gvwlNq9JH71_lt_U-nF0BXq9IoUfM9DNFK7EQF6NPqcphQJx0wtgil6pnfGStbDTbKBgYyut7LWxgb8eBtKhqGrvD3KS45mhLhK6pOAhj_piuohEY1daeNXq7rDwB

(Only applicable for extension issues) IDE Information

No response