docs: overhaul basic IO how-to

[NickCrews]

• removes a redirection using {{< include /_code/input_output_penguins.qmd >}}, just put the content there directly.
• organizes by type of destination: backend-native, file, in-memory
• merges the cross-database stuff into the basics so people find it, and flags it very explicitly so that when someone does ibis.postgres.connect().read_parquet() they don't get immediately frustrated. This should help with user stories such as Ibis drops or ignores timezones #11879
• talks about local vs remote filesystems explicitly
• I believe frames ibi's role as providing a unified API for IO, not for actually implementing it.
• moves pyarrow before pandas and polars and explicitly says it's our best-supported inmemory format.
• Actually walks through HOW to get eg parquet data into sqlite. The old multiple databases doc wasn't very useful for this.