v1.1.2: Tuned MCP surface + bugfixes

A release that makes SLayer easier for agents to discover and fixes two naming problems in query results.

Better MCP experience for agents. Every MCP tool description now fits the 2048-character limit Claude Code applies, so agents see the whole thing instead of a truncated one. The query-language reference now lives on the query schema fields where you write each value, and five new help topics (help.aggregations, help.transforms, help.time, help.joins, help.queries) cover the rest. inspect now returns the full text of any help.* memory. You can pass --always-load-query to slayer mcp / slayer serve (or set SLAYER_MCP_ALWAYS_LOAD_QUERY) to have Claude Code always load the query tool. When a measure name collides with a column, the error now suggests a free name.

Named joined-path dimensions keep their name. A dimension like {"name": "region_id", "expression": "customers.region_id"} is now keyed as orders.region_id and works in multi-stage queries and query-backed models. Stage schema errors now name the stage.

Readable auto-names for expression aggregates. Aggregates over date comparisons, nested aggregates or transforms no longer get garbled result keys. Breaking: some auto-names change, e.g. sum(customers.spend - 1) from orders is now orders.customers.spend_1_sum, and sum(ordered_at >= '2024-02-01') is now orders.ordered_at_2024_02_01_sum. If a saved query filters or orders by one of these keys, update it or set an explicit name.