Add LLGo runtime performance job

[zhouguangyuan0718]

Depends on #21. This is intentionally separate from the upstream-sync PR; after #21 lands, this PR diff reduces to the runtime job only.

The job:

• runs only when the xgo-dev/llgo release workflow dispatches a newly published tag and exact commit; ordinary pushes and pull requests do not trigger it;
• verifies that the dispatched tag exists in xgo-dev/llgo and resolves to the dispatched commit;
• compares Go, LLGo, and LLGo full LTO with Bent -N=10 -G;
• runs 12 pinned cases selected from a local benchmarks-100.toml survey against LLGo cf91afb98fca979c5195751554b67ec75f8c71d6;
• includes six cases with a measured LLGo/Full-LTO advantage and six stable same-range cases for optimization tracking;
• validates that every toolchain produced ten passing samples for every case;
• uploads raw Bent output plus a benchstat report;
• reports measurements without a hard percentage gate, since hosted-runner timing is noisy.

Local validation:

• go test ./...
• workflow YAML parse
• bash -n ci/llgo-performance/report.sh
• Bent initialization/listing for the checked-in configs
• report validation against 12 x 3 x 10 passing local samples