Remove --download-mode and the diff download path (#792) (#1049)

* Remove --download-mode and the diff download path (#792)

Patch content is now always fetched as per-file blobs. v5 has not
shipped stable, so the deprecated diff path goes before it does.

Core:
- delete patch/diff.rs (bspatch) and the qbsdiff dependency
- apply: drop the diff strategy, AppliedVia::Diff,
  PatchSources::diffs_path and the now-unused `uuid` parameter of
  apply_package_patch / apply_go_redirect
- blob_fetcher: drop DownloadMode, get_missing_archives and
  fetch_missing_sources; download_entries is blob-only. DIFF_ARCHIVE
  stays as the noun for the obsolete-archive sweep output
- client: drop fetch_diff
- package.rs: drop read_archive_filtered only; the other readers are
  used by the vendored and hosted backends
- cleanup: .socket/diffs is obsolete like .socket/packages and every
  file in it is swept
- telemetry: patch_fetched keeps download_mode, always "file"

CLI:
- remove --download-mode / SOCKET_DOWNLOAD_MODE outright (no alias,
  no warning), matching how #966 removes deprecated spellings
- fetch_stage stages blobs only; with every afterHash blob staged up
  front, apply's mismatch-blob top-up is dead and is removed
- repair downloads blobs only; the Downloaded event keeps
  details.mode, always "file"
- appliedVia is always "blob"

Tests: delete the diff e2e suites and diff/mode parse tests, add
removal parse tests, a cold-cache no-/diff-request assertion and a
stale referenced .socket/diffs archive sweep check.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

* Document the --download-mode removal for v5 (#792)

CLI_CONTRACT.md: drop the flag and env rows, note the removal, make
appliedVia "blob" only, and mark .socket/diffs obsolete in the GC
notes. migrating-to-v5.md: add the Retired spellings row and say
.socket/diffs archives are no longer read.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

* Update the GlobalArgs field count after removing download_mode (#792)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

* Drop stale references left by the diff path removal (#792)

Remove the deleted diff_created_file_e2e.rs from the bare-spawn ratchet
(its stale-entry check failed), point leftover comments at the blob-only
pipeline, and test repair's download messages with the blob noun.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

* Stop logging the archive uuid in a remove test assert (#792)

CodeQL flags the uuid as sensitive (cleartext logging); label the
entries A/B in the assert message instead.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

* Retrigger CI after a CodeQL SARIF upload infra failure (#792)

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

* Keep per-blob download failure reasons under --silent apply

With the diff download path gone, per-item blob failures print only via
format_fetch_summary, which is gated on quiet (--silent or --json). A
fatal staging failure under --silent then printed the generic error
without the hash and reason lines the old deferred-failure path showed
(Bugbot 4223475496). Print format_fetch_failures before the error when
the summary was held back; --json stays silent on stderr.

Co-Authored-By: Claude <noreply@anthropic.com>

* Drop --download-mode from the repair org-fallback test

The test came in from main with #1041 and still passed --download-mode
file, which this PR removes, so repair rejected the argument and CI
went red on the main merge. Repair always fetches per-file blobs now,
so the flag is simply dropped.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

* Drop removed download_mode field from target-ambiguity test

main's new in_process_target_ambiguity test (#1034) sets
GlobalArgs::download_mode, which this PR deletes, so the merge at
d06ce6f does not compile the cli test targets.

Co-Authored-By: Claude <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Author: mikolalysenko

Cargo.lock

@@ -42,7 +42,6 @@ aho-corasick = "=1.1.4"
 glob = "=0.3.4"
 toml_edit = "=0.25.12"
 once_cell = "=1.21.3"
-qbsdiff = "=1.4.4"
 rayon = "=1.12.0"
 tar = "=0.4.46"
 flate2 = "=1.1.9"
@@ -96,7 +95,7 @@ inherits = "release"
 lto = "thin"
 
 # Test-execution speed: `cargo test` builds dependencies with the dev
-# profile; at opt-level 0 the hash/compression/bsdiff hot loops are 10-100x
+# profile; at opt-level 0 the hash/compression hot loops are 10-100x
 # slower (the self_update fixtures gzip and sha256-hash the multi-MB debug
 # CLI binary per test: ~403s debug vs ~2s release). Workspace members are NOT matched by "*" — they stay
 # opt-level 0, so incremental compile speed, debugging, and llvm-cov line
@@ -107,7 +106,7 @@ opt-level = 1
 
 # The measured hot path gets full optimization: archive building
 # (flate2/zip over multi-MB payloads), SHA hashing in fixtures and in the
-# CLI's own verify paths, and qbsdiff bspatch in every apply test.
+# CLI's own verify paths.
 [profile.dev.package.sha2]
 opt-level = 3
 [profile.dev.package.sha1]
@@ -132,14 +131,6 @@ opt-level = 3
 opt-level = 3
 [profile.dev.package.zip]
 opt-level = 3
-[profile.dev.package.qbsdiff]
-opt-level = 3
-[profile.dev.package.bzip2]
-opt-level = 3
-[profile.dev.package.libbz2-rs-sys]
-opt-level = 3
-[profile.dev.package.suffix_array]
-opt-level = 3
 
 # macOS debug builds use unpacked split-debuginfo: every linked binary —
 # including each of the ~90 e2e test executables — pins its per-codegen-unit