Avoid redundant scans when extracting stage0 archives - #165819

…ns, r=Mark-Simulacrum

Avoid redundant scans when extracting stage0 archives

[`getnames()`](https://github.com/python/cpython/blob/b4e48a444ea02921ce4b701fe165e6cfd4cf5845/Lib/tarfile.py#L1991-L1995) calls [`getmembers()`](https://github.com/python/cpython/blob/b4e48a444ea02921ce4b701fe165e6cfd4cf5845/Lib/tarfile.py#L1981-L1989), which can scan the entire archive if its members have not already been loaded. Iterating over tar directly (https://github.com/python/cpython/blob/v3.10.12/Lib/tarfile.py#L2715-L2747) allows archive entries to be handled one at a time. Passing each `TarInfo` directly to [`extract()`](https://docs.python.org/3.10/library/tarfile.html#tarfile.TarFile.extract) also avoids name-based lookup, removing the preliminary scan and its associated repeated decompression.

With cached stage0 archives, `RustBuild.download_toolchain()` improved from 6.44 s to 3.50 s (medians of two runs per version). The rustc, rust-std, and cargo archives were extracted in parallel into fresh directories. Output hashes were verified separately.

Rollup of 6 pull requests

Successful merges:

 - #165778 (Bump stage0 to 1.101.0-beta.1)
 - #163077 (feat: support nested functions with `#[offload_kernel]`)
 - #165578 (Avoid `Hash::hash` when stable-hashing various types)
 - #165729 (tests: Add regression test for infinite recursive dereference)
 - #165819 (Avoid redundant scans when extracting stage0 archives)
 - #165822 (Remove unused compiler modules and types)

…ns, r=Mark-Simulacrum

Avoid redundant scans when extracting stage0 archives

[`getnames()`](https://github.com/python/cpython/blob/b4e48a444ea02921ce4b701fe165e6cfd4cf5845/Lib/tarfile.py#L1991-L1995) calls [`getmembers()`](https://github.com/python/cpython/blob/b4e48a444ea02921ce4b701fe165e6cfd4cf5845/Lib/tarfile.py#L1981-L1989), which can scan the entire archive if its members have not already been loaded. Iterating over tar directly (https://github.com/python/cpython/blob/v3.10.12/Lib/tarfile.py#L2715-L2747) allows archive entries to be handled one at a time. Passing each `TarInfo` directly to [`extract()`](https://docs.python.org/3.10/library/tarfile.html#tarfile.TarFile.extract) also avoids name-based lookup, removing the preliminary scan and its associated repeated decompression.

With cached stage0 archives, `RustBuild.download_toolchain()` improved from 6.44 s to 3.50 s (medians of two runs per version). The rustc, rust-std, and cargo archives were extracted in parallel into fresh directories. Output hashes were verified separately.

Rollup of 12 pull requests

Successful merges:

 - #165804 (cg_llvm: Use an invariant lifetime brand for LLVM values)
 - #165778 (Bump stage0 to 1.101.0-beta.1)
 - #165412 (Implement `core::ops::FnAddr`)
 - #165487 (Migrate built-in macros `pattern_type` & `view_type` to macros 2.0 that expand to internal syntax)
 - #165818 (do not lint `redundant_lifetimes` on items from external macros)
 - #163077 (feat: support nested functions with `#[offload_kernel]`)
 - #164795 (Extend support of lint `explicit_outlives_requirements` to checked type aliases)
 - #165578 (Avoid `Hash::hash` when stable-hashing various types)
 - #165729 (tests: Add regression test for infinite recursive dereference)
 - #165819 (Avoid redundant scans when extracting stage0 archives)
 - #165822 (Remove unused compiler modules and types)
 - #165824 (Add regression test for HRTB projection bound with TryFrom error)

Avoid redundant scans when extracting stage0 archives


try-job: dist-x86_64-msvc

Avoid redundant scans when extracting stage0 archives


try-job: dist-x86_64-msvc