Module eml_kernel 

EML learned functions for kernel subsystems.

Each model replaces a hardcoded heuristic with a trainable eml_core::EmlModel, falling back to the original fixed logic when the model has not yet been trained.

Models

Model	Replaces	Inputs	Outputs
GovernanceScorerModel	EffectVector L2 norm	5 dimensions	1 composite score
RestartStrategyModel	Fixed backoff delays	4 features	2 (delay, should_retry)
HealthThresholdModel	Fixed probe thresholds	3 features	2 (degraded, failed)
DeadLetterModel	Fixed retry policy	3 features	2 (delay, should_discard)
GossipTimingModel	Fixed gossip intervals	3 features	1 interval
ComplexityModel	500-line threshold	3 features	1 threshold

Structs

ComplexityModel

Learned complexity threshold replacing the fixed 500-line limit.

DeadLetterModel

Learned dead letter retry policy replacing fixed backoff and discard rules.

GossipTimingModel

Learned gossip interval replacing fixed timing constants.

GovernanceScorerModel

Learned governance scorer replacing the L2 norm on EffectVector.

HealthThresholdModel

Learned health check thresholds replacing fixed probe configuration.

RestartStrategyModel

Learned restart strategy replacing fixed backoff delays and retry decisions.

TickIntervalModel

Learned tick-interval recommender replacing the four-tier step-function in crate::weaver::WeaverEngine::recommend_tick_interval.