fix: db() reflects owned-state commits (read-your-writes)

[bplatz]

Problem

Closes #1330

Calling fluree.db(ledger_id) after committing through the owned-state API could return a previously cached GraphDb view that predates those commits — a read-your-writes violation.

The owned-state API (insert / update / transact(ledger, ...) and the transact(...).execute() builder) consumes a caller-supplied LedgerState, commits, and publishes to the nameservice, then returns a new LedgerState — but it never wrote that state back into the LedgerManager cache. A subsequent db() / ledger_cached() on a warm cache therefore served the pre-commit view.

The guarded builder path (fluree.stage(&handle).execute()) was never affected: it holds the ledger write lock across stage + commit and replaces the cached state via finalize_commit. The server transaction path (LocalCommitter) uses that guarded path, so server-side read-your-writes was already correct — this only affected the embedded owned-state API.

Fix

• Add Fluree::finalize_owned_commit, a shared post-commit tail for the owned-state methods that writes the committed state back into the cache (via sync_owned_commit_to_cache) before computing the indexing status and triggering background indexing. This also consolidates six duplicated post-commit tails into one.
• sync_owned_commit_to_cache is monotonic (skips when the cache has advanced past the new state) and only touches an already-cached handle — an uncached ledger loads fresh from the nameservice head on next access. On a refresh_index failure it evicts the entry so the next access reloads rather than serving stale state.
• Add LedgerManager::get_loaded_handle to fetch a cached handle without forcing a load.

Tests

• Drop the db_at_t workarounds in it_cyclic_bgp_probe and it_join_batched_overlay (the sites the issue called out).
• Add it_read_your_writes as a direct regression test.
• it_notify_incremental: in-process commits no longer leave the writer's own cache stale, so these now create a genuine cross-writer gap via a peer LedgerManager (the realistic scenario notify catch-up exists for).
• it_query_sparql_indexed: head-with-novelty views now take the production overlay fast path, which renders a decimal AVG as a precision-preserving string.

Verification

• Full fluree-db-api suite (2253 tests) + fluree-db-server / fluree-db-consensus tests pass.
• cargo fmt --all --check and cargo clippy --all --all-features --all-targets -- -D warnings clean.

Docs

Updated docs/getting-started/rust-api.md and docs/concepts/time-travel.md to state that a writer reads its own writes in-process, with eventual consistency remaining a cross-process concern.