fix(linked): support multiple operator schemas

[voltjia]

Summary

• Allow a linked binding's existing operator_schema field to contain either one schema string or a non-empty list of schema strings.
• Validate every listed schema against the same resolved provider library and dispatch key while loading that DSO only once.
• Preserve scalar manifest output for one schema and use a list only for multiple schemas.

Motivation

A linked composite can require several Dispatcher operators from one provider DSO. Expressing those exact contracts in the existing field avoids a new dependency graph, a new YAML key, or an imprecise high-level provider claim.

N/A - no linked issue.

Type of Change

• feat - new feature / new operator / new platform
• fix - bug fix
• perf - performance improvement (no behavioral change)
• refactor - code restructuring without behavior change
• test - adding or fixing tests only
• docs - documentation only
• build / ci - build system or CI configuration
• chore - tooling, formatting, or other non-code changes
• Breaking change

Platforms Affected

• CPU (WITH_CPU)
• NVIDIA (WITH_NVIDIA)
• Iluvatar (WITH_ILUVATAR)
• MetaX (WITH_METAX)
• Cambricon (WITH_CAMBRICON)
• Moore (WITH_MOORE)
• Ascend (WITH_ASCEND)
• PyTorch C++ bindings (WITH_TORCH)
• Build system / CMake / CI
• Python bindings / user-facing API

Smoke Test Result

No operator implementation or compiled target changes in this PR. The resolver path was exercised directly with the real vLLM provider DSO, which is the behavior changed here.

tests/test_resolve_linked_ops.py: 32 passed

# vllm/vllm-openai:v0.10.0, real vllm/_moe_C.abi3.so
SCHEMAS_VALIDATED=2
LOAD_LIBRARY_CALLS=1
RESOLVED_OPERATOR_SCHEMAS=2
RESOLVED_LIBRARIES=1
FORCE_LOAD_ENTRIES=1

Test Results on Supported Platforms

Platform	Affected	Build / Smoke Result	Full Result / Notes
NVIDIA	No	N/A - no backend source changed	Real _moe_C.abi3.so resolver gate passed
Iluvatar	No	N/A - not affected	N/A
MetaX	No	N/A - not affected	N/A
Cambricon	No	N/A - not affected	N/A
Moore	No	N/A - not affected	N/A
Ascend	No	N/A - not affected	N/A

Additional validation

Ruff 0.15.22 check and format: passed
git diff --check: passed
empty, non-string, blank, and duplicate schemas: rejected
one-element list manifest output: canonical scalar
multi-element list manifest output: ordered list

Benchmark / Performance Impact

N/A - configuration resolution only.

Notes for Reviewers

• This extends the value type of the existing operator_schema field; it does not add a new configuration key or dependency mechanism.
• Bare Tensor! remains the parseable representation for independent mutable aliases. The resolver retains its strict canonical schema comparison.
• All schemas in one binding intentionally share its single dispatch_key.