fix: consistent xsd:decimal/xsd:integer datatype across index and novelty (#1329)

[bplatz]

Summary

Fixes #1329: an xsd:decimal served from the binary index rendered with an empty @type ({"@value":"19.99","@type":""}) while the same value in novelty rendered as a bare string. Also fixes a sibling mislabel where indexed > i64 integers came back as xsd:decimal.

Root cause

OType::NUM_BIG_OVERFLOW names no single datatype from o_type alone — its arena holds both overflow xsd:integer (BigInt) and xsd:decimal (BigDecimal). So resolve_datatype_sid(o_type) returns None, and every flake/binding builder fell back to Sid::new(0, "") → empty @type. Novelty copies still carry o_type = XSD_DECIMAL, so they resolved correctly — hence the index-vs-novelty asymmetry.

A second, related issue: the late-materialized EncodedLit for the NumBig arena hardcodes dt_id = DECIMAL (it has no decoded value at creation time), so an indexed overflow integer materialized as xsd:decimal.

Fix

• New BinaryIndexStore::resolve_datatype_sid_for_value(o_type, &val) derives the datatype from the decoded FlakeValue (Decimal→xsd:decimal, BigInt→xsd:integer) when o_type can't name it. Wired into every flake/binding builder that has the decoded value in scope: binary_range (x2), binary_history, object_binding, fast_group_count_firsts (x2), block_fetch.
• binary_scan::matches_datatype_constraint is a pre-decode filter — it dropped indexed big numerics from datatype-constrained patterns (e.g. {"@value":"?p","@type":"xsd:decimal"}) while novelty copies passed. It now decodes to disambiguate only in the rare ambiguous case, keeping the common path decode-free.
• New FlakeValue::overflow_numeric_datatype_sid() recovers the datatype at materialization time (the EncodedLit can't be fixed at creation). Consulted before the dt_id lookup in both the query materializer and the API formatter; resolve_datatype_sid_for_value delegates to it too.

Because xsd:decimal is an inferable datatype, the consistent output is the bare-string form across crawl and tabular lanes (matching the existing convention). export.rs already had per-variant fallbacks, so it needed no change.

Tests

• jsonld_decimal_renders_consistently_across_index_and_novelty — crawl + tabular SPARQL + datatype-constrained pattern, asserting the indexed and novelty decimals render identically and both match a xsd:decimal-constrained query.
• indexed_overflow_integer_reports_xsd_integer_not_decimal — pins the late-materialization datatype recovery.

Both confirmed to fail before the fix and pass after. Full fluree-db-api suite, fluree-db-query/fluree-db-core unit tests, and clippy (all-features) are green.