[timtofan]

Depends on #471

Why

An idle connection reaching its configured read deadline is an expected lifecycle event. Counting that close as a decode error makes input error metrics misleading.

Architecture

The plaintext and stream reader loops recognize timeout errors and return normally. The change is local to input error classification; protocol decoding, routing, admission, and output ownership retain their existing boundaries.

Before / after

Before	After
An idle read timeout increments an input error counter	An idle read timeout closes the connection normally
A malformed or truncated stream is an input error	Malformed and truncated streams remain errors
Points parsed before the timeout are processed	Those points still flush normally

Validation

Go 1.27.0, vendored dependencies: build, vet, serialized full tests, and golangci-lint v2 with the upstream configuration. Regression tests cover idle plaintext, idle stream input, and genuine decode failures.