BUG: pricing resolver returns the aggregator route for equally priced models (test failing on main)

tests/test_pricing.py::test_resolves_common_bare_model_names fails on current main (v1.6.2, 52b1923):

    assert resolve_litellm_model("grok-4.5") == "xai/grok-4.5"
E   AssertionError: assert 'openrouter/x-ai/grok-4.5' == 'xai/grok-4.5'

Full run: 1740 passed, 1 failed.

Cause

LiteLLM 1.90.1 (the version pinned in uv.lock) lists grok-4.5 under three routes with identical pricing:

key in litellm.model_cost	litellm_provider	input	output
xai/grok-4.5	xai	2e-06	6e-06
openrouter/x-ai/grok-4.5	openrouter	2e-06	6e-06
perplexity/xai/grok-4.5	perplexity	2e-06	6e-06

In strix/report/pricing.py, resolve_litellm_model collects the matching keys, and when the prices tie it returns matches[0] from an alphabetically sorted list:

matches = sorted(key for key in model_cost if key.endswith(f"/{name}"))
...
if len(matches) == 1 or len(prices) == 1:
    return matches[0]

openrouter/... sorts before xai/..., so the aggregator route now wins over the provider's own. Nothing in the repo changed — the LiteLLM registry gained the aggregator entries.

Suggested fix

Among equally priced matches, prefer the canonical route: the one shaped exactly {litellm_provider}/{name}. An explicitly aggregator-qualified name such as x-ai/grok-4.5 should still resolve to the aggregator, since that is what the caller asked for.

I have this implemented and tested locally (1741 passed, ruff/mypy/pyright clean) and will open a PR referencing this issue.

Environment

• Strix main @ 52b1923 (v1.6.2)
• Python 3.12, litellm 1.90.1 (from uv.lock)
• Linux