Introduce "strict" mode for `bigint` setting

[nbenn]

Commit 641ec8b on branch dev for adbi adds experimental support for "strict" bigint options: "integer-strict" and "numeric-strict" both error on lossy data conversions (due to overflow or rounding).

Using "integer-strict" as default causes issues with tests:

• [ ] data_64_bit_numeric
• [ ] data_64_bit_numeric_warning
• [ ] data_64_bit_lossless
• [ ] arrow_read_table_arrow

Further issues might exist but currently do not reach the point where this issue comes up. Candidates include

• [ ] roundtrip_64_bit_numeric
• [ ] roundtrip_64_bit_character
• [ ] roundtrip_64_bit_roundtrip

[krlmlr]

Thanks for the heads-up, looks good!

We should expand the specs here. It will take some time because I want to inline the checks first. The skips in adbi are fine for now, except I don't see why arrow_read_table_arrow would fail.