Resolve package cycle around MissingServletRequestPartException

[jhoeller]

Since #27910 / #27948 web.multipart.MissingServletRequestPartException extends web.bind.ServletRequestBindingException which unfortunately creates a package cycle since the web.bind package depends on web.multipart within its binder implementations.

The root of the problem there is that MissingServletRequestPartException serves two rather different purposes: It is being thrown by RequestPartServletServerHttpRequest as a low-level API exception (which is why it lives in the web.multipart package), but then also used by RequestParamMethodArgumentResolver for handler method argument binding (for which it should actually live in web.bind). The proper solution would be to use a different exception type for the bind purpose, and to only let that one extend ServletRequestBindingException as of 6.0. Let's revisit this for M5.

[rstoyanchev]

It looks like originally MissingServletRequestPartException came as a result of differentiating 4xx vs 5xx exception related to multipart handling, see #13284 and related fix 56c8c69, and prior to that RequestParamMethodArgumentResolver used to raise the base class ServletRequestBindingException.

I think it would be still useful to have a single exception for a missing part, no matter where it was detected. RequestPartServletServerHttpRequest is under web.multipart because it relates to multipart processing and that's the lowest place where it is needed.

One alternative would be to revert the changes under #27948, i.e. make it extend ServletException again, and document more explicitly the reason it does not extend ServletRequestBindingException is that it can be raised at a lower level and lives in a different package. In retrospect, I suspect #27910 was more a question than an actual problem. This essentially answers the "if there is no good reason" question under #27910.