[rithikasridharbabu-2007]

Checklist

• The title and commit message(s) are descriptive.
• Small commits made to fix your PR have been squashed to avoid history pollution.
• Tests have been added for new features or bug fixes.
• API of new functions and classes are documented.

Description

Assigning arange or linspace to an existing contiguous container currently creates a full-size temporary because generators conservatively retain overlap protection. This adds an opt-in is_memory_independent trait, forwards it through xgenerator, and opts in arithmetic arange_generator specializations, used by both builders. Same-size assignments can then reuse existing storage.

Container assignment validates the shape before entering a generator’s noexcept assign_to. Unknown generators, including generators that read the destination, retain overlap protection. Forced-temporary mode is respected. Views and dynamically strided containers retain their current assignment paths: direct view assignment was slower than generating into a contiguous temporary in the local diagnostic.

Related to #2943. This is a focused first step for plain arithmetic arange/linspace assignments, rather than a complete solution for all generators or composed expressions.

Validation

• Nine relevant test executables passed: test_xbuilder, test_xassign, test_xarray, test_xtensor, test_xview, test_xrandom, test_xnoalias, test_xadapt, and test_xfixed. Automatic overlap detection was enabled with XTENSOR_FORCE_TEMPORARY_MEMORY_IN_ASSIGNMENTS=OFF.
• New regression tests cover zero data-buffer allocations for same-size arange/linspace assignment, resizing and empty output, an unknown generator reading the destination in reverse, self-concatenation, and dimension mismatch. The assignment suite passes 5 cases / 49 assertions.
• Three diagnostic runs on macOS arm64 with AppleClang 17, C++20 and -O3 -DNDEBUG showed the additional element-buffer allocation falling from 8 MiB to zero for 1,048,576 doubles. All generated values were checked. Allocation instrumentation and default-allocator timings were measured separately. Large container timings were close; no general runtime speedup is claimed.
• A separate forced-temporary build still allocated the expected temporary for ordinary assignment.

Remaining validation / draft status

• The complete combined test target does not compile on the unchanged baseline in this environment: masked-view tests trigger reference-binding and deleted-assignment errors in xtl::xmasked_value. Reproduced with xtl 0.8.0 and 0.8.2. Full-suite success is not claimed.
• A focused ASan/UBSan executable compiled, but its run produced no output and was interrupted. Sanitizer validation remains incomplete.

Opening as a draft for feedback on the trait and the container-only scope while these validation limits remain.