feat: make mainExpandBy and separatorExpandBy configurable options

[Theroxenes]

Summary

Web search highlights are expanded on both sides by expandHighlights to give
the model surrounding context, using hardcoded mainExpandBy = 300 /
separatorExpandBy = 150 (chars per side). This expansion can be the dominant
contributor to web search payload size.

Measured on a real search (local SearXNG + Firecrawl-compatible scraper +
CrossEncoder reranker), per source, holding the reranked chunks constant and
varying only the expansion params:

Expansion (main/separator)	Post-expansion bytes/source	Ratio vs. reranked chunks
300 / 150 (current default)	~4,200	~6.7×
100 / 50	~1,700	~2.8×

A ~58% per-source reduction, with highlight coherence roughly preserved.

This PR exposes the two parameters via config, following the existing
topResults / strategies / safeSearch convention (config-only — no env
var). Defaults are unchanged, so the change is non-breaking.

expandHighlights already accepts these parameters with the correct defaults;
this PR simply threads config values to the existing call site rather than
adding new machinery.

Changes

• types.ts: add mainExpandBy / separatorExpandBy to ProcessSourcesConfig
  (inherited by SearchToolConfig).
• tool.ts: destructure both from config with defaults (300/150) and pass them
  to the expandHighlights call.
• source-processing.test.ts: add coverage for default expansion, smaller →
  shorter highlights, larger → longer, zero → chunk-only, and the
  processor→expand path.

Change Type

• New feature (non-breaking change which adds functionality)

Testing

• npx jest src/tools/search — all existing + new tests pass.
• npx tsc --noEmit, npx eslint src/tools/search/ clean.
• Verified end-to-end against a real search pipeline: setting smaller expansion
  values produces correspondingly shorter highlights in the formatted output,
  with the core reranked passage preserved.