Reflexion: learn from failures that repeat #485

What

A Bot repeats a failure it has already hit: a broken tool is retried in a loop, and a failing model call is retried the same way. Nothing about that failure is kept, so the Bot never changes course.

Expected behavior

• When a tool call keeps erroring, or a model call fails several times in a row, the Bot writes itself one short, actionable note into its own memory ("the tool "X" has failed several times…").
• The note is written at most once per signature per day, so a loop cannot flood memory.
• It goes through the ordinary memory path: size-capped, secret-scanned, visible in the memory panel, removable with forget, and it never throws into the turn.
• config.reflexion: false turns it off.

Change scope

New core/src/host/reflexion.js (pure tracker + wording), core/src/host/memory.js (reflexionNote through the existing addEntry), core/src/host/turns.js (report failed tool calls and failed turns), tests core/test/reflexion.test.js + one case in core/test/memory.test.js, docs/subsystems/core-host.md.

Why

A Reflexion-style self-improvement loop (see the roadmap discussion), reusing the tested memory write path so there is no new prompt plumbing and no new failure mode.