feat(cli): support --model flag to override adapter model per run

[nextlevelshit]

Summary

• Adds a --model flag to wave run, wave do, and wave meta commands
• The flag overrides runtime.adapter.model from wave.yaml for a single invocation
• Per-persona model pinning takes precedence over the CLI flag
• The override is threaded through RunOptions → executor → adapter cleanly
• Invalid or empty model identifiers are rejected with clear error messages

Closes #198

Changes

• cmd/wave/commands/run.go — Added --model persistent flag to the run command, threaded into RunOptions
• cmd/wave/commands/do.go — Added --model persistent flag to the do command
• cmd/wave/commands/meta.go — Added --model persistent flag to the meta command
• internal/pipeline/executor.go — Extended RunOptions with ModelOverride field; executor applies override to adapter options when persona doesn't pin a model
• internal/pipeline/executor_test.go — Added comprehensive tests for override logic, per-persona precedence, and empty-override passthrough
• cmd/wave/commands/run_test.go, do_test.go, meta_test.go — Added flag registration tests
• specs/198-model-flag-override/ — Added specification, plan, and task artifacts

Test Plan

• Unit tests cover: flag parsing on all three commands, model override application, per-persona model pinning precedence, no-op when flag is absent
• All existing tests continue to pass (go test ./...)