Support E4M3B15 datatype (#765)

## Summary

- **Add `fp8_e4m3b15` datatype**: A software-defined FP8 type with 4
exponent bits, 3 mantissa bits, and bias=15 (max finite value: 0.9375).
Implemented entirely in software with no HW dependency, using
Triton-style bit manipulation through fp16 as intermediate for efficient
conversion.
- **Add mixed-precision accumulation for allreduce**: All allreduce
algorithm variants (packet, NVLS packet, fullmesh, RSAG zero-copy, and
others) now support a configurable `accumDtype` parameter, enabling FP8
inputs to be reduced in float16 or float32 for higher accuracy.
- **Propagate `accumDtype` through the full API**: The new parameter is
threaded from `Algorithm::execute()` → `NativeAlgorithm` → `KernelFunc`
→ dispatch → CUDA kernels, with `DataType::AUTO` as the default
(resolves to input dtype at runtime).
- **Add FP8 accumulation correctness tests**: New `test_fp8_accum.py`
validates that higher-precision accumulation produces results at least
as accurate as native FP8 accumulation across multiple algorithms and
sizes. Skipped on CUDA SM < 89 (pre-Hopper); runs on HIP/ROCm.
- **Add `test_fp8_accum.py` to CI**: Azure Pipeline `ut.yml` now runs
FP8 accumulation tests alongside existing pytests.
- **NCCL shim logging cleanup**: Migrated `printf`-style `WARN`/`INFO`
calls to streaming-style logging.

## Key files

| Area | Files |
|------|-------|
| New datatype + vector ops | `include/mscclpp/gpu_data_types.hpp` |
| Accumulation reduce helpers | `src/core/include/reduce_kernel.hpp` |
| Algorithm API (`accumDtype`) | `include/mscclpp/algorithm.hpp`,
`src/core/algorithm.cc` |
| Allreduce kernels | `src/ext/collectives/allreduce/*.cu` |
| Dispatch + common | `src/ext/collectives/include/allreduce/common.hpp`
|
| Python bindings | `python/csrc/algorithm.cpp`,
`python/mscclpp/_core/algorithm.py` |
| Tests | `python/test/test_fp8_accum.py` |
| CI | `.azure-pipelines/templates/ut.yml` |

## Test plan

- [x] CI passes on H100 (CUDA SM 90) — full FP8 E4M3 + E4M3B15
accumulation tests
- [x] CI passes on A100 (CUDA SM 80) — FP8 tests correctly skipped
- [x] CI passes on MI300X (ROCm) — FP8 tests run via HIP
- [x] Existing `test_mscclpp.py` tests continue to pass
- [x] NCCL shim builds and runs correctly with new `accumDtype` defaults

🤖 Generated with [Claude Code](https://claude.com/claude-code)

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

python/mscclpp/_core/algorithm.py

@@ -177,6 +177,7 @@ class Algorithm:
         nthreads_per_block=0,
         symmetric_memory: bool = False,
         extras: Optional[Dict[str, int]] = None,
+        accum_dtype: Optional[CppDataType] = None,
     ) -> int:
         """Execute the collective algorithm.
 
@@ -194,10 +195,14 @@ class Algorithm:
             nthreads_per_block: Number of threads per block (0 for auto-selection).
             symmetric_memory: Whether to use symmetric memory optimization (default: False).
             extras: Additional algorithm-specific parameters.
+            accum_dtype: Data type for accumulation during reduction. If None, defaults to
+                         the same as dtype. Use DataType.float32 for high-precision FP8 accumulation.
 
         Returns:
             The result code (0 for success).
         """
+        merged_extras = dict(extras) if extras is not None else {}
+        accum_dtype = accum_dtype if accum_dtype is not None else dtype
         return self._algorithm.execute(
             comm,
             int(input_buffer),
@@ -211,7 +216,8 @@ class Algorithm:
             nblocks,
             nthreads_per_block,
             symmetric_memory,
-            extras if extras is not None else {},
+            merged_extras,
+            int(accum_dtype),
         )
 
     def reset(self):