v0.100.0: burn by window, how fast the herd is burning

moshcode 0.100.0 tells you how fast the herd is burning, not just what it burned.

moshcode cost (and usage) answered one question: what did today cost, per session, with a total. The question that decides whether to kill something is a different one, and until now nothing answered it: how fast is it going right now, and is that a spike or the new normal. This release adds a burn block under the table, the same windows for every engine.

What changed:

• Under the session table, burn shows the last 1 min, 15 min, 1 hour, 4 hours, 8 hours and the report window, each with what the requests inside it cost, that cost per hour of window, and the runs behind it, plus the per-minute average over the window. A window longer than --since is left out rather than shown short. --json carries the rows as burn, each with the cost split by engine. --watch keeps the block.
• Every engine reader now keeps each request as a sample, on the engine's own clock and in the one usage shape moshcode already normalised everything into, so bucketing by time is one function over claude, codex, opencode, privacycode, qwen and aider rather than six kinds of arithmetic. An engine that priced its own requests keeps its number; the rest go through the rate card, and ~ still means estimated. Codex logs running totals, so a turn is the difference between two of them; aider stamps only the start of a run, so its whole run lands there.
• Claude Code writes a session's subagent transcripts under <session>/subagents/, and a workflow's agents one level deeper, all carrying the parent's session id. The reader stopped at the top of the project directory and never saw them. On a box running workflows that was about a quarter of the bill. They now fold into the parent's run, deduped by message id across the session, because the parent replays a subagent's messages when it takes the output back.
• The rate card had no entry for Claude Fable 5.1, so the prefix match landed on Claude Fable 5 and priced cache reads at $1.00 per million tokens. Fable 5.1 reads the cache at $0.25, and on an agent session cache reads are most of the tokens, so most of every Fable 5.1 estimate was wrong. claude-fable-5-1 and claude-mythos-5-1 now carry their own cache read rate.

Also in this release, from 0.99.1 on main: the claude engine's defaults ship ultracode off and the ultracode keyword trigger off (#509), the public agent pricing and help read $400 per agent-hour (#510), and the docs explain the native Chovy MCP connection and its revocation (#516).