Bug: Downloading multiple data tables is slow

[sergiopeixoto-seequent]

Bug Description

Retrieving multiple data tables from a Geoscience Object is slow.

Steps to Reproduce

` dhc = await object_api_client.download_object_by_id(object_uuid)
dhc_dict = dhc.as_dict()

data_client = object_api_client.get_data_client(manager.cache)

tables_coroutines = []
data_urls = []
for collection in dhc_dict["collections"]:
    if collection["name"] == "DerivedCPTReadings":
        for attribute in collection["distance"]["attributes"]:
            tables_coroutines.append(dhc.download_table(attribute["values"]))
            data_urls.append(dhc._urls_by_name[attribute["values"]["data"]])

sdk_start = time.perf_counter()
sdk_tables = await asyncio.gather(*tables_coroutines)
sdk_elapsed = time.perf_counter() - sdk_start
print(f"Download with SDK took {sdk_elapsed:.2f}")

async def _download_one(session: aiohttp.ClientSession, url: str) -> pa.Table:
    async with session.get(url) as resp:
        resp.raise_for_status()
        return pq.read_table(BytesIO(await resp.read()))

async with aiohttp.ClientSession() as session:
    non_sdk_start = time.perf_counter()
    tables = await asyncio.gather(*[_download_one(session, url) for url in data_urls])
    non_sdk_elapsed = time.perf_counter() - non_sdk_start
    print(f"Download without SDK took {non_sdk_elapsed:.2f}") 

`

Expected Behavior

Download with SDK took 0.99
Download without SDK took 0.99

Actual Behavior

Download with SDK took 10.30
Download without SDK took 0.99

Environment

• Windows 11
• Python 3.12.3
• evo-compute==0.0.1rc3
• evo-sdk-common==0.5.15

Acceptance Criteria

The speed is the same using the SDK or using raw requests

[tim-hamblin-seequent]

I had a look into this. From what I can tell download_table re-downloads the entire object definition from the server every time it reads a single parquet file, just to look up one URL — so importing x data files makes x -1 redundant object requests, wasting a lot of time. We could fetch the object once, reuse its URLs, and download the files concurrently over a kept-alive connection, drastically reducing the time (an example I tried increased it by 18x).