to_string and from_string have different behavior for TableReference

### Describe the bug

`to_string` simply concat components with `.`:
https://github.com/apache/arrow-datafusion/blob/02a470f6061cce8ee8e57f7af8a6a0e0ddc1571b/datafusion/common/src/table_reference.rs#L107-L121

but `from_string` will transform them to lowercase:
https://github.com/apache/arrow-datafusion/blob/02a470f6061cce8ee8e57f7af8a6a0e0ddc1571b/datafusion/common/src/utils.rs#L248-L257

`TableReference` is the most common table identifier for tables. But I don't think converting cases or considering quotes is necessary when the object name becomes `TableReference`. At this stage (i.e., after the initial plan is constructed from AST), those object names should be fixed.

This implicit conversion makes it difficult when handling complex table identifiers. Thus I propose to make `TableReference` as simple as possible and gather all those `to_lower_cases` and "quotes" before generating the very first `TableReference`.

### To Reproduce

```rust
#[test]
fn to_lower_case() {
let table_ref = TableReference::Full {
catalog: Cow::Owned("catalog".to_string()),
schema: Cow::Owned("schema".to_string()),
table: Cow::Owned("TABLE".to_string()),
};
let concatted = table_ref.to_string();
println!("concatted: {}", concatted);
let parsed = TableReference::parse_str(&concatted);
println!("parsed: {:?}", parsed.to_vec());
}
```

The output is
```
concatted: catalog.schema.TABLE
parsed: ["catalog", "schema", "table"]
```

### Expected behavior

```
concatted: catalog.schema.TABLE
parsed: ["catalog", "schema", "TABLE"]
```

### Additional context

(Not sure if this is "bug"...)

Start with datafusion/common/src/table_reference.rs and the referenced parsing logic in datafusion/common/src/utils.rs, then run the reproducing to_lower_case test using TableReference::parse_str. Compare the current to_string and from_string behavior and define tests covering case preservation and quoted or complex identifiers; done means the expected TABLE value survives the round trip.

Written by the indexing model from the issue text.