ir-gen: Fix ICE when reading an indexed ref mut array element

[Dnreikronos]

Closes #7602

Reading an element of a ref mut array argument, like self[i] inside an impl ... for [T; N] method, crashed the compiler. A ref mut argument is already a pointer in IR, and the argument gets stored into a local on entry, so by the time the index-read code runs the prefix is a pointer to a pointer to the array. That shape wasn't handled and fell into a todo!(). Writing through the same self[i] worked because the reassignment path already dereferences ref mut arguments.

The index-read path now peels off the extra pointer indirection before indexing, the way reassignment does, and the previously disabled assert_eq(self[i], default) line in the const-generics reassignment test is back on.

[fuel-cla-bot]

Thanks for the contribution! Before we can merge this, we need @Dnreikronos to sign the Fuel Labs Contributor License Agreement.

[cursor]

PR Summary

Medium Risk
Touches core IR indexing for references and pointers; behavior change is narrow but compiler codegen paths are sensitive.

Overview
Fixes a compiler ICE when reading an indexed element on a ref mut array (e.g. self[i] in impl ... for [T; N]). IR indexing now loads through extra pointer layers on the prefix so the value used for get_elem_ptr matches the reassignment path, which already handled ref mut correctly.

Unsupported non-array/slice pointer targets no longer hit todo!(); they return an internal compile error with a span. The const-generics reassignment e2e test re-enables assert_eq(self[i], default) that was blocked on this bug.

^{Reviewed by Cursor Bugbot for commit b7bcec1. Bugbot is set up for automated code reviews on this repo. Configure here.}

[Dnreikronos]

@ironcev, can we re-trigger the cla pipe here and on the other PRs that I have open?

[codspeed-hq]

Merging this PR will not alter performance

✅ 25 untouched benchmarks

---

_{Comparing Dnreikronos:fix/index-read-ref-mut-array-7602 (b7bcec1) with master (cf225be)}