Effort level can now be adjusted per turn based on permission layers

What

• A new turnEffort value, derived from the active permission layers, now feeds into the effort-resolution logic used across compaction, the /effort command's output, the ${CLAUDE_EFFORT} prompt template substitution, and agent session metadata, alongside the existing model default and any explicit override.
• The effort-resolution function gained a turnEffort option that, when the model has a defined capability lookup, takes priority in choosing the effective effort level, and it now also affects the effort suffix text shown to users (e.g. "with X effort").
• Subagent (Task) tool calls now pass their own turnEffort, computed from the subagent's permission layers, into the permission checks used for command-execution and model resolution.

Why Allows the effort level to vary within a session based on context, such as which permission layer is active for a given turn or subagent, rather than being fixed for the whole session.