feat: add --no-expand flag to vsearch and query commands

[cryptomonkey84]

Adds a --no-expand flag to vsearch and query commands that skips query expansion, avoiding the need to download and load the 1.28GB qmd-query-expansion-1.7B GGUF model.

Problem

On low-RAM machines (8GB), the expansion model download gets OOM-killed before completing, making vsearch and query completely unusable. The 300MB embedding model works fine — it's specifically the query expansion step that's the blocker.

Fixes #158. Related: #147, #155 (same family of resource exhaustion issues).

Changes

• Added --no-expand boolean CLI flag
• When set, store.expandQuery is temporarily overridden to return [], skipping expansion in both vectorSearchQuery and hybridQuery
• Original function restored via try/finally
• Help text updated for both commands

Testing

qmd vsearch --no-expand "test query"    # Works without 1.28GB model
qmd vsearch "test query"                 # Unchanged behaviour
qmd query --no-expand "test query"       # Works without expansion model

Raw vector search still returns relevant results (57-63% similarity scores in testing). Expansion improves recall but isn't essential.