feat: add .mempalaceignore support + fix ChromaDB telemetry spam

[Matt-GPT-6]

Summary

Two focused fixes in one PR:

1. .mempalaceignore file support (closes #102)

Users can now create a .mempalaceignore (or .mempalace-ignore) file in any project directory to exclude files from mining — using the exact same syntax as .gitignore.

Why this matters: Issue #102 reported that mempalace mine scoops up junk files. While .gitignore is already honored, many files that should be excluded from memory (credentials, drafts, large data files) shouldn't necessarily be in .gitignore. A separate ignore file gives users fine-grained control without modifying their project's git config.

How it works:

• .mempalaceignore is always applied — even when --no-gitignore is passed (so users can skip gitignore but still respect mempalace-specific exclusions)
• Supports the full .gitignore syntax: globs, directory-only rules, negation (!keep-this.py), anchored patterns, ** wildcards
• Works at any directory depth, just like nested .gitignore files
• Both .mempalaceignore and .mempalace-ignore filenames are accepted
• Both filenames are auto-excluded from mining (added to SKIP_FILENAMES)
• Refactored GitignoreMatcher → IgnoreFileMatcher with a backwards-compatible alias so existing imports and tests don't break

2. Fix ChromaDB/posthog telemetry error spam

Every CLI command printed Failed to send telemetry event ClientStartEvent: capture() takes 1 positional argument but 3 were given to stderr.

Root cause: ChromaDB 0.6.x calls posthog.capture(user_id, event_name, properties) with positional args, but posthog >= 7.0 changed capture() to keyword-only args. The env var ANONYMIZED_TELEMETRY=False doesn't prevent the broken call from firing.

Fix: Neutralize posthog.capture early in __init__.py before ChromaDB initializes. MemPalace is a local-only tool — upstream telemetry is not useful.

Test plan

• All 20 existing tests pass (no regressions)
• 5 new tests added for .mempalaceignore behavior:
  • Basic file and directory exclusion
  • Works even with respect_gitignore=False
  • .mempalace-ignore (hyphenated) variant
  • Negation patterns (!keep.csv)
  • Overrides files that .gitignore would allow
• Manual verification: mempalace status, mempalace search produce clean output with no telemetry errors
• 25/25 tests pass on Python 3.13, macOS ARM64

Made with Cursor

[bensig]

Thanks! .gitignore support already landed in #78 which covers most of this. A separate .mempalaceignore is tracked in #102. The telemetry fix is a good catch though — could you open a focused PR for just that piece?

[PaTiToMaSteR]

@bensig if scans the code as it is, consumes 3 times the tokens compared with GREP. .mempalace-ignore to only scan *.md make it incredibly great, run a test performs much more! we have:

1. MCP layer consumes more than tool execution: mempalace search "" is way cheaper than the MCP layer
2. GREP performs better than the mempalace search, that consume a lot of tokens
3. for MD files or non-code knowledge seems to be incredibly great

Either there's a test to confirm that performs better for searching code lines, or we whitelist only *md files coz adding this MCP greatly consumes A LOT more on top.