fix:/prefer the use of column identity over ordinals

[johnbrk]

Does your PR solve an issue?

Fixes #4210

Is this a breaking change?

No, just makes query_as! behave more like query_as

[barry3406]

Nice catch — the ordinal-vs-name mismatch in query_as! has bitten a few people.

One thing I noticed: this only switches to column-name lookup for the (true, ColumnType::Exact) branch, but the Wildcard, OptWildcard, and (false, _) arms still use #i (the ordinal). If the column order can differ at runtime (the exact scenario in #4210), those branches would still break. Was that intentional, or should they also use stringify!(#column_ident)?

Also minor: quote::format_ident!("{}", ident) is a no-op when ident is already an Ident — you can pass &ident directly into the quote! interpolation and stringify!(#ident) will work the same way.

[abonander]

This was by design because we didn't want to do MxN hashmap lookups when reading a resultset, and it's one of the major reasons we try to discourage SELECT * (the other being that you should only select the columns you're actually going to read to avoid wasting bandwidth and server resources). I think this should just be more clearly documented (and maybe warn/error for top-level SELECT *).

However, I think we can get the best of both worlds with a change to how we read rows. Since the shape of the resultset doesn't change while you're reading it, we can resolve column names to ordinals once at the beginning of the resultset and use the ordinals when reading the rows. query_as could also take advantage of this.