Add an auxiliary MTR capital gains elasticity alongside the retention-rate elasticity

[MaxGhenis]

Makes the auxiliary elasticity convention native, so downstream analyses stop hand-rolling response machinery in either convention.

Context

#1803 made gov.simulation.capital_gains_responses.elasticity the retention-rate elasticity (Agersnap–Zidar / CenTax convention) on the hardened MTR measurement. Some analyses also report the marginal-tax-rate convention as an auxiliary comparison — PolicyEngine/uk-cgt-paper#7 documents one hand-rolling exactly this with the pre-#1803 nested-branch pattern, and PolicyEngine/uk-equalising-cgt#1 documents another approximating it with a constant conversion that inverts sign under the ≥2.90 semantics.

Change

• New parameter gov.simulation.capital_gains_responses.mtr_elasticity (default 0): elasticity of realisations with respect to the marginal tax rate itself, typically negative.
• capital_gains_behavioural_response applies it through the previously-orphaned relative_capital_gains_mtr_change variable (dlog MTR on measure_capital_gains_mtrs, 0.001 floor).
• Setting both elasticities nonzero for a period raises a ValueError naming both parameters — an ambiguous double response fails loudly before any early exit.
• The shipped retention path is untouched: existing tests pass unmodified.

Tests: exact hand-computed MTR factor, both-set guard, default zeros, recalculation regression; full CGT YAML suite passes.

🤖 Generated with Claude Code