Rename remaining `Registry` identifiers to `Index` for consistency with new index protocol

[consideRatio]

AI generated thing I figure we should do, but I don't want to bundle into another PR but instead do in isolation.

Background

After landing the index protocol (#288), the codebase uses three different names for what is now consistently called "the index" in ARCHITECTURE.md:

• index — used in ARCHITECTURE.md, the new env::index::IndexEnvironment{,Async}, the new project::index_entry::IndexEntryProject, and the index_resolver field name.
• Registry — used as the public type-parameter name and several enum variants/types in core/src/resolve/combined.rs.
• registry — used in the lockfile wire format (Source::Registry { registry } → [sources] registry = "...").

Three names for one concept is a comprehension trap, and the gap is widening as new code adopts Index*.

Inventory

A. core/src/resolve/combined.rs — pure rename, no wire-format impact

RegistryResolver               (type param)            → IndexResolver
RegistryProjectStorage         (type param)            → IndexProjectStorage
RegistryError                  (type param)            → IndexError
CombinedResolverError::Registry        (variant)       → ::Index
CombinedProjectStorage::RegistryProject(_)             → ::IndexProject
CombinedProjectStorage::CachedRegistryProject(_)       → ::CachedIndexProject
CombinedIteratorState::ResolvedRegistry(_)             → ::ResolvedIndex
field `index_resolver`         (already correct, no change)

Plus matching updates in:

• doc comments (/// - registry_resolver is what will typically be hit when using non-resolvable IRIs, /// A resolver for whatever is considered a central project registry, // Finally try the centralised registry if neither file/remote gave anything useful, "registry-project checksum_canonical_hex failed; …")

B. Lockfile wire format — core/src/lock.rs:560, core/src/lock_tests.rs:304

Source::Registry { registry: String }

serializes to TOML as:

[[projects.sources]]
registry = "..."

Renaming this variant + key would be a lockfile schema break — every existing sysand-lock.toml containing such a source would fail to parse. Two options:

1. Defer. Keep Source::Registry { registry } as-is for now; do (A) only. The naming inconsistency between combined.rs and lock.rs becomes "registry is the wire word, index is the conceptual word" and gets a one-line doc-comment to that effect on the variant.
2. Bundle with a lockfile-version bump. Rename to Source::Index { index }, increment LockfileVersion, and add a deserializer migration (or a hard error pointing at sysand lock --upgrade).

Recommend option 1 unless a lockfile-version bump is on the near roadmap for an unrelated reason.

Why now

ARCHITECTURE.md is now the authoritative vocabulary source and uses "index" exclusively. Recent renames (HTTPIndexEnvironment → IndexEnvironment, IndexedRemoteProject → IndexEntryProject, file moves to env/index.rs / project/index_entry.rs) reinforce that. combined.rs is the only remaining surface where a reader sees "Registry" and has to mentally translate to "index".

Out of scope

• Pre-existing direct-HTTP types in resolve/reqwest_http.rs (HTTPProjectAsync etc.) — separate cleanup, kept as-is per recent review.
• The commands/lock.rs lockfile writer logic — only the variant name changes if (B) is taken.

Acceptance

• (A) all Registry* identifiers in core/src/resolve/combined.rs renamed to Index*
• doc comments and log messages in the same file updated
• decision recorded on (B): defer (with one-line doc on the variant) or bundle (with lockfile version bump)
• cargo test --workspace --all-features --exclude sysand-py clean