fix(trail): record_count/summary/export include unflushed buffered records (#150)

[rajfirke]

What does this PR do?

In buffered mode, record_count, summary(), and export() all queried only the backend, giving a stale view of the trail whenever records were sitting in the write buffer:

• record_count returned 0 (or the flushed count) until the next background flush
• summary() reported stale provenance/freshness/source aggregates — a MISSING provenance in the buffer was invisible
• export() silently omitted pending writes entirely

Fix:

WriteBuffer gains two helpers:

• __len__ — delegates to pending so callers can use len(buffer)
• pending_records — returns a thread-safe snapshot of the buffer contents

ContextTrail changes:

• record_count adds self._buffer.pending to the backend count without flushing (read-only)
• summary() appends self._buffer.pending_records to the backend snapshot before aggregating
• export() calls self._buffer.flush() before reading — the export must be complete; a non-flushing approach would return an inconsistent snapshot here

Also cleans up two leftover issues from the [1.1.0] offset addition:

• Removes the stray # Added offset parameter to truly handle datasets of any size without hard caps comment in trail.py
• Renames test_context_trail_query_supports_offset → test_paginated_mock_trail_returns_distinct_pages to reflect that it tests PaginatedMockTrail, not ContextTrail.query()

Checklist

• Tests added/updated for the change
• ruff check src/ tests/ passes
• ruff format --check src/ tests/ passes
• mypy src/provena/ passes
• pytest passes with no failures
• CHANGELOG.md updated (if user-facing change)

Related Issues

Closes #150

---

Note: PR #161 covered the same fix. After a week without the requested revisions, this replaces it with the corrected implementation. Closing #161 with a note.