feat(memory): add token-friendly brief retrieval

[Grivn]

Summary

• add optional --brief output to recall and search
• add a Unicode-safe --excerpt-chars limit, defaulting to 240 characters
• add mnemon show <id> for retrieving one complete active insight
• keep existing/default JSON output unchanged for compatibility
• document the new retrieval flow in English and Chinese

Brief responses contain only the retrieval fields needed to choose a result (id, excerpt, category, score/confidence) plus a single detail_command. JSON is compact and does not HTML-escape the command placeholder. --brief and --verbose are mutually exclusive.

The public output remains JSON rather than making the current TOON Working Draft a protocol dependency.

Token benchmark

For ten deliberately long results with the o200k tokenizer:

• current compact output: 10,576 tokens
• brief output: 732 tokens (93.1% reduction)
• brief plus one full show: 1,819 tokens (82.8% reduction)

Verification

• focused brief/show and Unicode truncation tests
• make test
• 156/156 CLI E2E assertions
• go test -race ./cmd/memory
• git diff --check

Closes #43.