update for 2.29 | NKI beta 3

[jlonge4]

Issue #, if available:

N/A

Description of changes:

1. nisa.memset strict type matching — Changed value=0.0 to value=0 in nki_matmul_fully_optimized_. NKI 0.3.0 enforces that the value literal must match the destination tensor's dtype; 0.0 (Python float64) is a type mismatch against lhsT.dtype.
2. nisa.tensor_copy — removed deprecated dtype parameter — The dtype keyword argument has been removed from nisa.tensor_copy in NKI 0.3.0. Removed from three call sites across nki_matmul_basic_, nki_matmul_tiled_, and nki_matmul_hoist_load_. The cast is now expressed entirely through the destination tensor's declared dtype.
3. nki_matmul_hoist_load_ — corrected res_sb allocation dtype — The res_sb SBUF buffer was incorrectly declared as nl.float32 while the intent was to store the final output-typed result. Changed to result.dtype to align with the other kernels and correctly express the PSUM→output dtype cast via the destination allocation.

Testing:

Please see detailed unit test requirements in the CONTRIBUTING.md

• The change is covered by numeric check using nki.baremetal
• The change is covered by performance benchmark test using nki.benchmark
• The change is covered by end-to-end integration test

Pull Request Checklist

• I have filled in all the required field in the template
• I have tested locally that all the tests pass
• By submitting this pull request, I confirm that my contribution is made under the terms of the MIT-0 license.