Compress complete Linux runtime archives across large decoder binaries

docs/LINUX-SPEECH-0.1.6.md

@@ -128,3 +128,12 @@ Ignored raw evidence is in `speech-auto-61afa08/relocated-verification.json` and
 `relocated-native-cpu.log` under the WSL validation cache.
 Linux GPU execution, shared-service packaged speech/Google retrieval and public
 archive delivery remain open. Archive construction is now underway.
+
+The first complete archive passed all 60 extracted file/permission records and
+both installed-layout self-tests, but at 109314948 bytes it exceeds Gitea's
+100-MiB upload limit. SHA-256:
+`9ae7fa9a203c6471cfd6bdd82ac5bf81683813c15fe652595976344fbbf96084`.
+It is retained as an unpublished candidate. XZ packaging now uses a 64-MiB
+dictionary (preset 9) to compress duplicate decoder/runtime code across large
+ELF files; client decompression needs roughly 65 MiB. All 21 package tests
+passed after this compression change. No runtime files are removed.

tools/package_linux.py

@@ -221,7 +221,9 @@ def package(root, executable, runtime, media, output, llama_license=None, compre
         temporary_path = Path(temporary.name)
     try:
         records = []
-        with temporary_path.open("wb") as raw, (lzma.LZMAFile(raw, "w", preset=6) if compression == "xz" else gzip.GzipFile(filename="", mode="wb", fileobj=raw, mtime=0)) as compressed, tarfile.open(fileobj=compressed, mode="w|") as archive:
+        # A 64-MiB dictionary retains duplicated decoder/runtime code across
+        # large ELF files. Decompression needs roughly 65 MiB on the client.
+        with temporary_path.open("wb") as raw, (lzma.LZMAFile(raw, "w", preset=9) if compression == "xz" else gzip.GzipFile(filename="", mode="wb", fileobj=raw, mtime=0)) as compressed, tarfile.open(fileobj=compressed, mode="w|") as archive:
             for name, (path, expected_record) in sorted(selected.items()):
                 safe_name(name)
                 regular(path, path.parent)