• This PR implements the core component of Fee Estimation via Fee rate Forecasters tracking issue #30392, introducing a new fee rate forecasting module.
  The primary addition is a ForecasterManager that coordinates multiple forecasting strategies to be able to provide better transaction fee rate predictions.

Key Changes

1. Addition of Fee Rate Forecasting Utility Structures

   • Forecaster abstract class, serving as the base class for all fee rate forecasters.
   • ForecastResult (the response from a Forecaster) with all metadata.
   • ForecastType enum, identify and distinguish forecasters.

2. Introduce FeeRateForecasterManager class

   • Adds the FeeRateForecasterManager class, responsible for managing all fee rate forecasters, including CBlockPolicyEstimator, which is now a forecaster.
   • Updates the node context to hold a unique pointer to FeeRateForecasterManager instead of CBlockPolicyEstimator.
   • Changes CBlockPolicyEstimator instance from unique_ptr to a shared_pointer, allowing it to be registered in the validation interface without requiring explicit unregistration during shutdown (current master behaviour). Registering for CBlockPolicyEstimator flushes also get a shared_pointer.
   • Exposes a raw pointer of CBlockPolicyEstimator through FeeRateForecasterManager for compatibility with estimateSmartFee, friends, and node interfaces.
   • Modifies CBlockPolicyEstimator to inherit from the Forecaster class and implement the pure abstract classes

3. Introduce MempoolForecaster class

   • Adds the MempoolForecaster, which forecast the fee rate required for a transaction to confirm as soon as possible. It generates a block template and returns the 75th and 50th percentile fee rates as low-priority and high-priority fee rate forecasts.
   • Implements caching for the last fee rate forecast, using a 30-second rate limit to avoid frequent block generation via the block assembler.

The FeeRateForecasterManager now includes a method ForecastFeeRateFromForecasters. This method polls forecasts from all registered forecasters and returns the lowest value. This behavior aligns with discussions and research outlined in this post. But the primary aim of the FeeRateForecasterManager is to be able to perform sanity checks and potentially apply heuristics to determine the most appropriate fee rate forecast from polled forecasters, for now it just return the lowest.

This PR also add unit tests for the classes.

In Draft because it depends on #33218