Support Overriding the Regex Size Limit

**Describe the bug**

#### Code

```rust
use arrow::array::StringArray;
use arrow::compute::regexp_is_match_utf8;

fn main() {
let array = StringArray::from(vec![Some("foo")]);
let regex_array = StringArray::from(vec![Some(r#"\w{1000}"#)]);
let ret = regexp_is_match_utf8(&array, ®ex_array, None).unwrap();
println!("{:?}", ret);
}
```

#### Error

```rust
called `Result::unwrap()` on an `Err` value: ComputeError("Regular expression did not compile: CompiledTooBig(10485760)")
stack backtrace:
0: rust_begin_unwind

```

#### Reason

https://github.com/rust-lang/regex/blob/8856fe36ac7dc37989e6ffb26b5fc57189bae626/src/builders.rs#L50-L57
```rust
impl Default for Builder {
fn default() -> Builder {
let metac = meta::Config::new()
.nfa_size_limit(Some(10 * (1 << 20))) // Default 10MB
.hybrid_cache_capacity(2 * (1 << 20));
Builder { pats: vec![], metac, syntaxc: syntax::Config::default() }
}
}
```

#### Solution

Replace `Regex::new(pattern.as_str())` with `RegexBuilder::new(pattern.as_str()).size_limit(regex_size_limit)`, and `regex_size_limit` can be obtained as a function argument or an environment variable.

If the community approves of this solution, I will implement it.

**To Reproduce**

**Expected behavior**

**Additional context**

Start at the compute::regexp_is_match_utf8 entry point shown in the reproduction and trace how patterns are compiled. Compare the current Regex::new path with the issue's proposed RegexBuilder size_limit approach, then verify that the reported large pattern no longer fails and that the chosen configuration path is covered by regression testing.

Written by the indexing model from the issue text.