Tracking: vendor every Maven shape through the jvm planner and retire the legacy single-pom backend #971

[agent] Filed by the scheduled architecture audit routine (ecosystems and formats). Register: discussion #560 register.

Kind: tracking. Source: review Part 5.7 ("JVM: two backends for one ecosystem"); register E26.

Problem

Vendored Maven has two backends, chosen by root-pom shape (verified on main @ 9c43dfc):

• vendor_maven sends reactors, Gradle builds and mixed roots to the v5 jvm planner. It sends a lone single-module pom.xml to the legacy vendor_maven_single, because Detected::shape maps (Some(Single), false) to Shape::Other ("Single-POM builds retain the legacy backend", jvm/mod.rs#L4).
• The planner already handles single-module poms. Shape::Mixed plans a single-module pom beside a Gradle build with maven_reactor::plan_with_config. A throwaway test, run twice on 9c43dfc, planned a lone CRLF single-module pom that detect classifies as Other:
  • the plan succeeded with no warnings;
  • re-planning wrote nothing;
  • unplan restored pom.xml byte for byte and deleted .mvn/maven.config.
• The two models differ in a user-visible way.
  • Legacy appends a <repository> socket-patch-vendor-<uuid> → file://${project.basedir}/.socket/vendor/maven/<uuid> and keeps the unsuffixed GAV. A warm ~/.m2, an earlier repository or mirrorOf * therefore serves the pristine jar (local_cache_shadow_warning is always emitted).
  • The planner pins <base>-socket.<hex8>, a GAV that only the committed .socket/vendor/maven2 tree serves.
• The split costs code at every layer:
  • the JVM orchestrator (vendor_maven_jvm, acquisition, metadata, materialisation; maven_repo.rs#L693-L1866,`` ~1,170 lines) lives inside the legacy backend's file;
  • entries are relabeled "jvm" (#L1344), which forces "maven" | "jvm" aliases at the CLI revert match, path.rs#L92, redownload.rs#L71-L96 (+ #L183, #L319, #L352) and the hosted takeover;``
  • three artifact roots (maven/<uuid>, maven2, gradle);
  • a second declares_modules (#L2019,`` cause of Vendored Maven refuses a single-module EAR pom as a multi-module aggregator because two declares_modules disagree #716);
  • the legacy_mixed_root bridge (#L711-L726).``

Symptoms (legacy same-GAV wiring): #263, #274 (single-module half), #622, #716.

Target design

One vendored JVM backend: jvm::detect returns a planner shape for every root that has a pom.xml or a Gradle build. A single-module pom is a reactor of one. The legacy maven_pom_repository ledger kind remains revert-only (adapted at load, the way legacy_mixed_root already keeps old roots working until they are reverted).

Checklist (in order)

• Move the v5 JVM vendor orchestration and Maven acquisition out of maven_repo.rs into vendor/jvm #972 (child 1): move the v5 JVM orchestration and the shared Maven acquisition out of maven_repo.rs into vendor/jvm/ (mechanical, no behavior change).
• Decide: vendor single-module Maven poms through the suffixed-version jvm planner and retire the same-GAV <repository> wiring #973 (child 2, decision): route single-module poms through the planner, and decide how existing legacy entries migrate.
• Child 3 (after the decision): detect routes Single to the planner; the forward legacy path (vendor_maven_single, maven_prelude, build_repo_edit writer, the comment-stripping declares_modules, local_cache_shadow_warning) is deleted, and revert_maven_opts' legacy arm stays.
• Child 4: one ledger ecosystem name; drop the "jvm" ↔ "maven" aliases (old ledgers adapted at load).
• Child 5: --maven-config becomes a vendor / scan --mode vendored flag (CLI surface; may fold into C34 / Decide: warn on and then remove scan --apply/--vendor, and whether --vex stays embedded #966).

Dependencies

Child 1 should land after #876, which edits fetch_registry_bytes in the same file. Blocks the VendorBackend table (#959) for Maven, and E24's revert engine (one fewer revert mechanism).