v0.24.0 — every tool reads the whole edit

Fixed

• The reporting tools were still blind to a gap-based storyline. v0.23.0
  fixed the parser and the media tools; an audit of the remaining tl.clips
  call sites found nine read-only tools that still walked the spine alone.
  On the #23 project export_edl and export_csv produced an empty export,
  analyze_pacing answered "No clips to analyze", list_markers and
  list_keywords found nothing, and filter_by_role reported no matches —
  each silently, with no hint the edit was in lanes. Same defect as #23, one
  surface out. Timeline.all_clips() is the base view (spine plus every
  connected clip, including the ones with no media, because a caption still
  carries markers, keywords and a role), and media_clips() is now the
  subset of it that has media. list_markers, list_keywords,
  export_csv and filter_by_role read all_clips(); export_edl and
  analyze_pacing read media_clips(), because an EDL row names a source a
  caption does not have and a caption is not a cut to average into pacing.
  list_markers also missed markers on connected clips in spine-based
  projects — a marker on a connected title was never listed.
• The preview placed connected clips by their raw offset. A clip under a
  gap whose clock starts at 3600s was drawn against a 3600s origin and
  landed four seconds early on the #23 project. Connected clips now render
  at their timeline_start, and the origin heuristic ignores elements the
  parser has already normalised (so a gap-based storyline, whose spine gaps
  establish 0 and are not kept as clips, gets an origin of 0).
• detect_media_silence reported spans of 0.00s. ffmpeg routinely puts
  a silence start tens of microseconds before a clip's out-point; clamping
  that to the used window left a span too short to print and impossible to
  act on, since split_clip cannot cut inside a frame.
  map_silence_to_timeline takes a min_mapped floor and the handler
  passes one frame duration. The default is 0, so other callers are
  unchanged.

Added

• Timeline.all_clips() — every clip carrying the edit, positioned on the
  timeline, spine or lanes.

Install / upgrade: uvx fcp-mcp-server picks it up on the next run; pip install -U fcp-mcp-server otherwise.

Tests: 1781 pass / 7 skip on the mcp floor, 1782 / 6 on mcp 2.x, 1757 / 31 with FCP_MCP_INDEX=off.