Add exclude_top_levels option to energy budget corrector

[mcgibbon]

The constant_temperature total-energy-budget correction applies a vertically- and horizontally-uniform air-temperature increment to enforce total-energy conservation. It currently adds that increment to every vertical level, including the thin model top level. In the 4°/daily ERA5-only baseline this introduces a positive bias in the topmost air temperature level. This PR adds an option to exclude the top N levels from receiving the increment, so the bias can be tested/removed without changing the conserved global magnitude.

Changes:

• fme.core.corrector.atmosphere.EnergyBudgetConfig: add exclude_top_levels: int = 0 field (with a non-negative __post_init__ check); it only restricts which levels receive the per-level temperature addition, the global magnitude of the correction is still computed across all levels.

• fme.core.corrector.atmosphere._force_conserve_total_energy: accept exclude_top_levels and skip the topmost N air_temperature levels (level 0 is the model top) when applying the increment; raise if it exceeds the number of levels. Threaded through AtmosphereCorrector.__call__.

• fme.core.corrector.test_atmosphere: excluded top level receives no increment while remaining levels get the same per-level magnitude as the no-exclusion case; negative-value and too-many-levels error cases.

• Tests added

• If dependencies changed, "deps only" image rebuilt and "latest_deps_only_image.txt" file updated

[mcgibbon]

Validating this change on the 4°/daily v2 ERA5-only baseline with the energy corrector and exclude_top_levels: 1 (excludes the model top level, level 0):

• beaker: https://beaker.org/ex/01KW2D1V2F4FY85R3N7M0GFWEV
• wandb: https://wandb.ai/ai2cm/ace/runs/bre6554p

Tests whether excluding the top level removes the positive air_temperature_0 bias the all-levels corrector introduced, without reintroducing energy-budget drift.