fix: Better casts to string for binary floats/doubles

[nj1973]

This PR is to work around problems validating floating point values where some engines return all digits and others return scientific notation, for example:

1.23456781E+007         │ 12345678.1

Changes in the PR:

• Use SQL Server FORMAT() instead of CAST() for real/float columns. In my tests that ensures we do not end up with scientific notation
• Use Teradata TO_CHAR('TM9') instead of CAST() for binary_float/double columns. In my tests that ensures we do not end up with scientific notation
• Use Oracle TO_CHAR('TM9') instead of CAST() for binary_float/double columns. In my tests that ensures we do not end up with scientific notation but we still get some unpredicatable values due to the lossy nature of binary floats. For this reason the column remain excluded from Oracle integration tests
• Reduces code duplication by importing the original Ibis _cast() function and calling that when not in one of our workarounds

Problems not fixed by this PR:

• Hive float and double remain problematic
• Oracle binary_float/double remain problematic

[nj1973]

/gcbrun

[nehanene15]

Looks great! I like the idea of importing sa_fixed_cast() to clean up the code