feat: multi-valued numeric DataValue and schema flag (PR1/2 of #281)

[mosuka]

Summary

• Add DataValue::Int64Array / Float64Array plus the multi_valued: bool flag on IntegerOption / FloatOption. Lucene semantics: array values per document, range queries match if any value satisfies the predicate (constant scoring), single → multi auto-wraps, multi → single is rejected (no silent truncation).
• Wire the new variants and the schema flag end-to-end through DataValue plumbing, type inference, type coercion, proto, REST / gRPC / MCP, and all six language bindings, with English and Japanese documentation refreshed.
• Lexical-store internals (doc_values Single/Multi split, BKD multi-point registration, range query "any match" evaluation) are deferred to PR2 so the storage contract changes in isolation.

What changed

Core (laurus)

• DataValue::Int64Array(Vec<i64>) and Float64Array(Vec<f64>), accessors, From<Vec<i64>> / From<Vec<f64>>, and DocumentBuilder::add_int64_array / add_float64_array.
• IntegerOption.multi_valued / FloatOption.multi_valued flags (default false, #[serde(default)]).
• infer_from_array now returns the new variants with multi_valued: true instead of "not yet supported".
• coerce_to_integer / coerce_to_float branch on multi_valued. Multi-valued accepts arrays, single values, bool, and parseable text (auto-wraps); single-valued rejects arrays explicitly.
• coerce_to_vector also accepts Int64Array / Float64Array so REST/JSON callers can still pass pre-computed embeddings as plain numeric arrays to vector fields.

Protocol & server

• New Int64ArrayValue / Float64ArrayValue proto messages added to the Value oneof. IntegerOption / FloatOption proto gain multi_valued.
• laurus-server::convert::document round-trips the new variants; convert::schema round-trips the flag; gateway JSON converters recognise multi_valued on input and emit array values on output. laurus-mcp::convert handles them too.

Bindings

• laurus-{python,nodejs,wasm,ruby,php}::Schema::add_integer_field / add_float_field accept a multi_valued parameter (default false). Each binding's data_value_to_* converter emits arrays using its language's idiomatic type.
• laurus-cli::create prompts for multi_valued on numeric fields; output::format_data_value and data_value_to_json render the new variants.

Documentation (EN + JA)

• concepts/schema_and_fields — type-inference table now lists arrays, plus a new "Multi-valued numeric fields" section that documents the auto-wrap / reject behaviour and the constant-score range-query contract.
• laurus-cli/schema_format — TOML multi_valued documented.
• laurus-server/grpc_api — IntegerOption / FloatOption field lists updated.
• laurus-{python,nodejs,wasm,ruby,php}/api_reference — signatures plus one-line descriptions of multi_valued.

Out of scope (PR2)

• FieldDocValues Single/Multi split.
• BKD multi-point registration for the same (field, doc) key.
• RangeScorer constant scoring with "any value matches" semantics.

Until PR2 lands, multi-valued integers/floats round-trip through the engine surface (DataValue, schema, ingest, retrieve via stored fields) but range queries against multi-valued fields are not yet evaluated correctly — that requires the storage-layer changes in PR2.

Test plan

• cargo fmt --check
• cargo clippy -p laurus -p laurus-server -p laurus-cli -p laurus-mcp -p laurus-python -p laurus-nodejs -p laurus-wasm --all-targets -- -D warnings
• cargo test -p laurus -p laurus-server -p laurus-cli -p laurus-mcp — 698 unit tests + 10 dynamic-schema integration tests pass.
• New unit tests in engine::type_inference::tests::infer_integer_array_to_int64_array / infer_float_array_to_float64_array.
• New integration test dynamic_auto_adds_int64_array_field in dynamic_schema_test.
• markdownlint-cli2 \"docs/src/**/*.md\" \"docs/ja/src/**/*.md\" — 0 errors.
• mdbook build docs and mdbook build docs/ja succeed.

Refs #281