Implement Live Monitoring System (`ferret watch`)

[apiad]

Description:

Currently, Ferret is a "post-mortem" analysis tool. It records span data to BeaverDB, and users must run ferret analyze after the fact to see what happened. This model is insufficient for debugging real-time performance issues, such as stuck processes, deadlocks, or live memory leaks.

To support real-time observability, we need a system that can visualize the current state of the application, including active (unfinished) spans and system-level metrics (CPU, RAM, Threads).

Proposed Solution:

1. Event Sourcing Model: Transition the storage model from logging just "Finished Spans" to logging "Lifecycle Events" (SPAN_START and SPAN_END). This allows a watcher to reconstruct the "currently open" call stack.
2. System Telemetry: Introduce a background thread in the Profiler that logs system metrics (CPU, Memory, Active Tasks) at a fixed interval.
3. TUI Dashboard: Create a new CLI command ferret watch using Textual that connects to the BeaverDB in read-only mode, tails the log stream, and visualizes the live state.

Tasks:

• Data Model Updates (ferret/models.py):
  • Define a generic LogEntry model that can represent different event types: SPAN_START, SPAN_END, and METRIC.
  • Update SpanModel to support being serialized as partial events.
• Profiler Logic (ferret/core.py):
  • Update Profiler.start() (and __enter__) to immediately write a SPAN_START event to the non-blocking queue.
  • Update Span.end() to write a SPAN_END event.
  • Implement SystemSampler: A daemon thread that wakes up every 1s, captures psutil stats and threading.active_count(), and writes a METRIC event.
• The Watcher App (ferret/tui.py):
  • Create a Textual application.
  • Live Tailer: Implement a utility that reads BeaverDB from the last known offset, yielding new records as they appear.
  • State Reconstruction: Maintain an in-memory tree of "Open Spans" based on the Start/End events received.
  • UI Layout:
    • Live Stack: A Tree widget showing currently running functions and their elapsed time.
    • Metrics: Sparklines or numeric indicators for CPU/RAM.
    • Recent Activity: A scrolling table of the most recently finished spans.

Acceptance Criteria:

• ferret watch launches a TUI dashboard.
• When the profiled script sleeps or performs a long task, the function appears in the "Live Stack" view immediately.
• When the function finishes, it disappears from the "Live Stack" and appears in the "Recent Activity" table.
• System metrics (CPU/RAM) update in real-time (approx. every second).