perf: reduce scan time on large lockfiles

[sonukapoor]

Performance review of the hot scan path identified several opportunities to reduce wall-clock time, particularly for large monorepos with many transitive findings. Reported context: a Discord engineer noted the tool is "within acceptable range but slower than I'd like" on a ~600-repo monorepo.

Critical (biggest impact)

• Parallelize getVuln() detail fetches in scanner.ts — done in PR perf: parallelize network I/O in scanner to reduce scan time on large lockfiles #646. Gatsby cold cache: 28.4s → 12.7s (2.2x faster).
• Pre-warm packument cache before remediation loop in scanner.ts — done in PR perf: parallelize network I/O in scanner to reduce scan time on large lockfiles #646.

Important

• Replace spread-array accumulation in npm-lock-graph.ts — Set accumulators, O(E×V) → O(E). Done in PR perf(npm-lock-graph): use Set accumulators for edge lists #652.
• Cache pre-frozen node objects at construction time in npm-lock-graph.ts — done in PR perf(npm-lock-graph): pre-freeze node objects at construction time; remove redundant uniquePathArrays #654.
• Build a Map<string, PackageRef> by name before the remediation loop to replace linear packages.find() scans — done in PR perf(remediation): replace linear packages.find() with Map lookup for O(1) access #653.

Minor

• Collapse multiple findings.filter() passes into one in scanner.ts
• Capture new Date().toISOString() once before the cache-write loop
• Remove redundant uniquePathArrays() in pathsFor() — done in PR perf(npm-lock-graph): pre-freeze node objects at construction time; remove redundant uniquePathArrays #654.