breakdown scorer to modular files

[kaushikmitr]

This pull request introduces several new helper modules and refactors the SLO-aware router in the scheduling framework to improve configurability, prediction handling, and internal data management. The main changes include adding support for environment-based configuration, modularizing prediction and selection logic, and making the SLO context store thread-safe.

Key changes:

Configuration and Environment Variable Support

• Added config.go to provide environment-variable-driven configuration for SLO-aware routing, including weights, selection strategies, and exploration parameters. This makes the router highly configurable without code changes.

Prediction and Selection Logic Modularization

• Introduced prediction.go to encapsulate prediction generation, validation, and SLO headroom calculation for candidate pods, improving code clarity and separation of concerns.
• Added helpers.go with composite scoring and weighted random/max selection logic for pod selection, supporting new composite strategies and selection modes.

Request Header Parsing

• Added headers.go with utility functions for robust parsing of float and boolean headers from requests, including error handling for invalid or missing values.

SLO Context Store Refactor

• Refactored the SLO context store to use a thread-safe sync.Map instead of a plain map, and updated all accessors to use Load, Store, and Delete methods for concurrency safety.

SLO Request Context Enhancements

• The SLORequestContext now pre-initializes maps for prefix cache scores and prediction results, and records the request received timestamp for improved tracking and metrics. [1] [2]