[clang] [analyzer] Refine invalidation caused by `fread` (PR #93408)

[Balazs Benics via cfe-commits]

steakhal wrote:

Checkout the new implementation and the added "weird" fread tests.
FYI unfortunately our store interferes a bit, as you will see in the last test (`test_unaligned_start_read`) when the store does not purge the previous binding when we have an overlapping write to the store.
Consequently, when reading, we can read the old value - which was partially overwritten by the escape/invalidation. 

https://github.com/llvm/llvm-project/pull/93408