performance issue when batch fetch bars

[zillionare]

omicron >= 2.0.0.a35

when fetch one day's minute bars of all stock/index, _batch_get_persisted_bars will cost almost 120 secs in debug mode (53 seconds almost if not in debug).

Time consumed at:

1. read content from http request and unzip, almost 20 seconds
2. convert csv to dataframe, 3.7 secs if using 'c' engine and 1.x secs if 'pyarrow' engine is picked
3. group and convert dataframe to numpy structured array, 50 secs.
4. influxdb will cost almost 20 secs.

we need decrease time cost by both 1 and 3 to seconds.

for 1, maybe we can try:

1. use httpx instead of aiohttpclient
2. let pandas to do the unzip

for 3, no idea at all.