GeezerVison

As I was working on gh-120754 I noticed I kept adding more members and copying out individual members from the fstat call, and that it may be simpler / easier to just stash (and invalidate) the whole stat result rather than individul members. This is preparatory work for

1. Avoid calling isatty on open for regular files (Resolving Avoid calling isatty() for most open() calls #90102)
2. Reduce system calls by making more members available (Helping implement Speed up open().read() pattern by reducing the number of system calls #120754)

One important note, and why the member is called stat_atopen is that the values should only be used as guidance / an estimate. With individual members copied out this was also the case. While it's common for a file to not be modified while python code is reading it, other processes could interact with it and code needs to handle that. Two examples of this that I've come across: It is possible to change an fd so isatty result changes (see: gh-90102 and GH-121941) and a fd which is opened in blocking mode may have an ioctl used on it to change it to non-blocking (see: gh-109523). The general class of bugs here are commonly called time-of-check to time of use (TOCTOU, https://en.wikipedia.org/wiki/Time-of-check_to_time-of-use)

Given how common some specific patterns are (ex. Path().read_bytes()) it is still worthwhile to optimize those (Ex. disabling buffering results in a over 10% speedup in that case, GH-122111). The existing codepaths treated this correctly as far as I can tell.

This PR is a portion of GH-121593 which is being split up into smaller, hopefully easier to review chunks. Not calling isatty for regular files makes a small but measurable perf improvement for every "open and read whole regular file" python does.