[Feat] Optimize ASU task manager with lock-free slots

[AooooooA-C]

Purpose

This PR replaces the mutex-protected unordered_map in TaskManagerBase with a fixed-size slot table based on open addressing and atomic state transitions. The goal is to reduce global lock contention.

Modifications

The original implementation:

std::mutex mu_;
std::unordered_map<TaskId, std::shared_ptr<Context>> tasks_;

is replaced by a fixed-size slot array:

std::vector<Slot> slots_;

Each slot stores:

state
task_id
ctx

Test

1. Create and enter the build directory

cd /YOURPATH/unified-cache-management
mkdir -p build && cd build

2. Configure CMake

cmake .. \
  -DBUILD_UCM_ASU=ON \
  -DBUILD_UNIT_TESTS=ON 
  -DBUILD_UCM_STORE=OFF \
  -DDOWNLOAD_DEPENDENCE=ON \
  -DRUNTIME_ENVIRONMENT=ascend

3. Build the asu.test target

cmake --build . --target asu.test -j$(nproc))

4. Run the tests

./ucm/transport/kv/asu/asu.test --gtest_also_run_disabled_tests --gtest_filter='TransportTaskManagerTest*:ClientTaskManagerTest*:TaskManagerBaseBench*:SlotTaskManagerBaseBench*'

5. Test Results

Unit tests for SlotTaskManagerBase verifying submit/get/remove/getAll correctness:

Benchmark: TaskManagerBase (mutex + unordered_map) vs SlotTaskManagerBase (lock-free slots + IndexPool)

[nrj868]

Nice