Namespace Qavren.Edge.Ingestion.Onnx

Classes

EdgeChunkTokenizer: IChunkTokenizer over SP2's IEdgeTokenizer (spec 8.1, 11.2). It holds ONLY the IEdgeTokenizer: CountTokens(ReadOnlySpan<char>) forwards to CountTokens(ReadOnlySpan<char>), which SP2 forwards verbatim to Tokenizer.CountTokens at its defaults and which takes a span rather than returning an index, so no normalised string can leak through it.

OnnxIngestionBuilderExtensions: The one extension Qavren.Edge.Ingestion.Onnx publishes (spec 11.2).

ResourceMonitorThrottle

IIngestionThrottle over SP2's Read() (spec 10.2). The table below is evaluated top to bottom, first match wins, and every halving is clamped to MinBatchSize:

`LastPressure == Critical`	pause, `memory:critical`
`AvailableMemoryBytes < 48 MiB`	pause, `memory:floor`
`Thermal == Critical`	pause, `thermal:critical`
`Thermal == Serious`	batch / 2, delay 250 ms, `thermal:serious`
`LastPressure == Moderate`	batch / 2, `memory:moderate`
`AvailableMemoryBytes < 96 MiB`	batch / 2, `memory:warn`
`IsLowPowerMode == true`	batch = min(batch, 4), delay 100 ms, `power:low`
otherwise	proceed at the configured batch

ResourceMonitorThrottleOptions: What ResourceMonitorThrottle can be told to ignore, and how far it may shrink a batch (spec 10.2). Every default here changes behaviour.