sync dev -> main

…recated"

This addresses #1685 (`institutional_holders`) and also `get_earnings_dates()`.

Pandas issue is found here:
pandas-dev/pandas#53767
and the change in code here:
https://github.com/pandas-dev/pandas/blob/5cedf87cccd77c7b4b6aaa64bfec98b32b512f68/pandas/io/html.py#L1238

As for legacy Python 2.7 support: `io.StringIO` seems to be supported in
the versions I tested. See https://docs.python.org/2/library/io.html

Fix pandas FutureWarning: "Passing literal html to 'read_html' is deprecated"

Add cookie & crumb to requests. Involves several changes:
- fetch cookie & crumb, obviously.
- two different cookie strategies - one seems to work better in USA, other better outside.
- yfinance auto-detects if one strategy fails, and switches to other strategy.
- cookie is stored in persistent cache folder, alongside timezones. Refetched after 24 hours.

To have this work well with multithreading (yfinance.download()) requires more changes:
- all threads share the same cookie, therefore the same session object. Requires thread-safety ...
- converted data class to a singleton with "SingletonMeta":
 - the first init() call initialises data.
 - but successive calls update its session object - naughty but necessary.
- thread locks to avoid deadlocks and race conditions.

Add cookie & crumb to requests