Add EmptyStore, which can be used as a mocker for StoreV1.
Optimize CacheStore to support independent Lookup operations.

…#3)

…itch (#5)

* [feat] add shard_data_dir config while `true` as default value
* [opt] setup SpaceLayout with GlobalConfig
* [feat] shard data file with its hash prefix

Add FakeStore, which allows for stubbing of Store components, enabling end-to-end testing of UCM's performance limits.

* [Bugfix] unlink shm file at exit

* [Opt] update impl of SpaceLayout::StorageBackend

…#10)

* [opt] Increase the default cache buffer number
* [fix] Fixed the issue of Scheduler role failing to initialize PCStore

# Purpose

In a layerwise scenario, CacheStore calls the asynchronous stream API to
execute a Load task. Before the asynchronous task completes, the
task-related resources cannot be reclaimed. Therefore, we need to extend
the lifecycle of the task context to the stream synchronization point.

# Modifications 

Place the task context being asynchronously transmitted into a holder,
and release the holder after the asynchronous task synchronization
point.

# Test

CI passed with new added/existing test.

# Purpose

Add test and example for **LayerWise** and **ChunkBlock** feature on
`CacheStore`

# Modifications 

- Unit test in cc:
`ucm/store/test/case/cache/cache_trans_manager_test.cc`
- Example in py: `ucm/store/test/e2e/cache_layer_chunk_test.py`

# Purpose

When data is shuffled, the first N characters of the Block's hash are
used as subdirectories. The current implementation is fixed at 8,
meaning there can be a maximum of 16^8 subdirectories, which is too
large for storage.

# Modifications 

A configuration option `data_dir_shard_bytes` is provided, allowing
users to configure the character length of subdirectories. The default
value is `4`, meaning there can be a maximum of 16^4 = 65536
subdirectories.

# Test

CI passed with new added/existing test.

# Purpose
PipelineStoreBuilder supports dynamic registration. It also includes two
built-in pipelines:
- `Cache|Empty`
- `Cache|Posix`

…a lookup. (#644)

# Purpose
Currently, the metadata lookup logic performs existence checks on all
Block Keys in the provided sequence. However, Blocks can only be reused
on a strict prefix match. This means when a Block is missing in the
sequence, checking subsequent Blocks is unnecessary and adds latency to
the API call.

# Modifications 
Introduce a new strict prefix lookup interface `LookupOnPrefix`. This
interface stops checking as soon as it encounters a non-existent Block
and returns the maximum index of existing Blocks, avoiding unnecessary
lookups and reducing latency.

Task need to be released after done to release reference on shm buffer.