perf(collector): parse /proc/loadavg without heap allocation

[GreyPikachu]

Description

This pull request optimizes /proc/loadavg parsing by replacing std::istringstream with zero-allocation std::string_view token scanning and std::from_chars.

Motivation

std::istringstream allocates a heap buffer and incurs locale-dependent stream overhead on every collection cycle. Because MetricsCollector::Load() runs on every tick, eliminating dynamic string allocation preserves cache locality and reduces allocator pressure in the sampling hot path.

Changes

• Replaced std::istringstream in MetricsCollector::Load() with a localized std::string_view whitespace scanner and std::from_chars.

Verification

• Verified parsing against standard /proc/loadavg formats (e.g. 1 2 3 1/2 42).
• Existing core and integration tests pass without errors.