refactor: Fix remove_unnecessary_refs to ignore pointers

[ahomescu]

Change the behavior of remove_unnecessary_refs to not remove dereferences of pointers because those are generally required.

[fw-immunant]

Code changes look reasonable, but is there an example to look at to better understand the motivation for this change? This refactoring pass runs in CI for the testsuite--are there cases that were broken there and now fixed?

[kkysen]

LGTM, too.

[ahomescu]

Code changes look reasonable, but is there an example to look at to better understand the motivation for this change? This refactoring pass runs in CI for the testsuite--are there cases that were broken there and now fixed?

That's right, we have remove_unnecesary_refs enabled in CI now and there are a bunch of places where it failed, e.g.

error[E0308]: mismatched types
   --> src/xzlib.rs:263:17
    |
263 |         *have = (have).wrapping_add(ret as ::core::ffi::c_uint);
    |         -----   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ expected `u32`, found `*mut u32`
    |         |
    |         expected due to the type of this binding
    |
    = note:     expected type `u32`
            found raw pointer `*mut u32`

error[E0599]: no method named `expect` found for raw pointer `*mut Option<unsafe extern "C" fn(*mut c_void, *const i8, ...)>` in the current scope
     --> src/xpath.rs:18203:18
      |
18202 | /             (crate::src::globals::__xmlGenericError())
18203 | |                 .expect(
      | |                 -^^^^^^ method not found in `*mut Option<unsafe extern "C" fn(*mut c_void, *const i8, ...)>`
      | |_________________|

and more.

I think that &*x is always fine to elide when x is a reference, but never when it's a pointer (correct me if I missed a case).

I had a separate patch for the test suite to disable that transform, but might as well fix it instead.