Super fast extended tests and improved planning speed linux (#21084)

This makes sqlite extended finish in 5 minutes:
https://github.com/apache/datafusion/actions/runs/23362665318/job/67969547959

Currently on main it takes 20 minutes (and a month ago it would take two
hours 🤯)

---------

Co-authored-by: Andrew Lamb <andrew@nerdnetworks.org>

Author: blaginin

.github/workflows/extended.yml

@@ -173,10 +173,10 @@ jobs:
           ref: ${{ github.event.inputs.pr_head_sha }} # will be empty if triggered by push
           submodules: true
           fetch-depth: 1
-      - name: Setup Rust toolchain
-        uses: ./.github/actions/setup-builder
-        with:
-          rust-version: stable
+      # Don't use setup-builder to avoid configuring RUST_BACKTRACE which is expensive
+      - name: Install protobuf compiler
+        run: |
+          apt-get update && apt-get install -y protobuf-compiler
       - name: Run sqllogictest
         run: |
           cargo test --features backtrace,parquet_encryption --profile ci-optimized --test sqllogictests -- --include-sqlite

datafusion/common/src/utils/mod.rs

@@ -40,7 +40,7 @@ use std::cmp::{Ordering, min};
 use std::collections::HashSet;
 use std::num::NonZero;
 use std::ops::Range;
-use std::sync::Arc;
+use std::sync::{Arc, LazyLock};
 use std::thread::available_parallelism;
 
 /// Applies an optional projection to a [`SchemaRef`], returning the
@@ -923,10 +923,15 @@ pub fn combine_limit(
 ///
 /// This is a wrapper around `std::thread::available_parallelism`, providing a default value
 /// of `1` if the system's parallelism cannot be determined.
+///
+/// The result is cached after the first call.
 pub fn get_available_parallelism() -> usize {
-    available_parallelism()
-        .unwrap_or(NonZero::new(1).expect("literal value `1` shouldn't be zero"))
-        .get()
+    static PARALLELISM: LazyLock<usize> = LazyLock::new(|| {
+        available_parallelism()
+            .unwrap_or(NonZero::new(1).expect("literal value `1` shouldn't be zero"))
+            .get()
+    });
+    *PARALLELISM
 }
 
 /// Converts a collection of function arguments into a fixed-size array of length N