feat(benchmarks): add quota command for Model Proxy spend (#1167)

Bump kagglesdk to 0.1.37 and expose the new GetModelProxyQuotas endpoint
as `kaggle benchmarks quota`.

The command reports AI inference spend in USD, one row per refill period
(daily and monthly). This is distinct from the top-level `kaggle quota`,
which reports weekly GPU/TPU accelerator hours, so it lives under
`benchmarks` alongside `auth` and `init` — the other commands that hit
the Model Proxy.

Remaining is derived as total - used and clamped at $0.00 so an overage
reads as zero rather than a negative balance, matching the existing
quota_view_cli behavior.

Also extracts the 404/403 friendly-error mapping out of
_fetch_model_proxy_env into a shared _translate_model_proxy_error
helper, since the quota endpoint has the same beta-access and
phone-verification failure modes.