Use only after the caller knows prefill/input processing has completed.
remainingOutputTokens represents the maximum future output still expected
for this active request. safetyMarginTokens retains extra headroom until
final release. Both values are non-negative safe integers.
Optional progressive reconciliation applied alongside cumulative usage.
Use only after the caller knows prefill/input processing has completed.
remainingOutputTokensrepresents the maximum future output still expected for this active request.safetyMarginTokensretains extra headroom until final release. Both values are non-negative safe integers.