[wip]introduce chunksize

[qyh111]

Purpose

Add chunksize
What this PR does / why we need it?

Modifications

原先的接口

# 一个block的一个offset对应于HBM上的一个tensor
def load(self, block_ids: List[str], offset: List[int], dst_tensor: List[torch.Tensor]) -> Task:
def fetch_data(self, block_ids: List[str], offset: List[int], dst_addr: List[int], size: List[int]) -> Task:

现在需要改成

# 一个block的一个offset对应于HBM的n个tensor/n个address，这n个tensor不能单独dump，需要进行聚合
def load(self, block_ids: List[str], offset: List[int], dst_tensor: List[List[torch.Tensor]]) -> Task:
def fetch_data(self, block_ids: List[str], offset: List[int], dst_addr: List[List[int]], size: List[int]) -> Task:

或者说不改原来的接口，新增一个dump_data_batch的接口

struct Shard {
        Type type;
        Location location;
        std::string block;
        size_t offset;
        std::vector<uintptr_t> address; //一个offset对于多个地址
        size_t length; // n个tensor的总大小
        size_t owner;
        std::shared_ptr<void> buffer;
        std::function<void(void)> done;

Dump时在posix_queue里面将多个address聚合到预先申请的buffer中

shard.buffer = device->GetBuffer(shard.length);
    if (!shard.buffer) {
        UC_ERROR("Out of memory({}).", shard.length);
        return Status::OutOfMemory();
    }
    auto hub = shard.buffer.get();
    auto dAddr = new std::byte*[shard.address.size()];
    auto hAddr = new std::byte*[shard.address.size()];
    for (size_t i = 0; i < shard.address.size(); i++) {
        hAddr[i] = (std::byte*)hub + i * shard.length / shard.address.size();
        dAddr[i] = (std::byte*)shard.address[i];
    }
    auto status = device->D2HBatchSync(hAddr, const_cast<const std::byte**>(dAddr), shard.address.size(), shard.length / shard.address.size());

Does this PR introduce any user-facing change?

Test

How was this patch tested?