feat(runtime): Update pollrate behavior and add tests for it - #890

Why

The tracer previously capped poll_rate to a maximum of 5 seconds,
silently resetting any higher value back to 1. This prevented customers
from benefiting from coarser poll rates (e.g. 60 s, 120 s) to reduce
observability cost.

Additionally, GC metrics were collected using gc.get_count() and
gc.get_threshold(), which do not reflect actual GC activity. The correct
source is gc.get_stats(), which tracks cumulative collections,
collected, and uncollectable counters per generation — consistent with
the OpenTelemetry CPython semantic conventions
(cpython.gc.collections, cpython.gc.collected_objects,
cpython.gc.uncollectable_objects) and the
OpenTelemetry Python contrib system_metrics instrumentation,
both of which source these metrics from gc.get_stats().

Finally, the host agent timeout detection used a fixed 60-second window,
causing spurious disconnects when poll_rate exceeded 30 seconds.

What

• options.py: Replaced MAX_POLL_RATE = 5 with a VALID_POLL_RATES
  list [1, 5, 10, 20, 30, 60, 120, 180, 240, 300, 360, 420, 480, 540, 600].
  set_poll_rate() now rounds to the nearest valid value instead of
  rejecting anything outside {1, 5}.

• runtime.py: Switched GC collection from gc.get_count() /
  gc.get_threshold() to gc.get_stats(). Reports per-generation deltas
  for collections, collected, and uncollectable (keys
  gc.collections0-2, gc.collected0-2, gc.uncollectable0-2).

• host.py: Timeout window is now max(60, poll_rate * 2) seconds to
  avoid false resets at high poll rates. Added _send_heartbeat() — a
  lightweight HEAD request sent when no metrics/spans/profiles are
  produced in a cycle, keeping last_seen current between polls.

Testing

Unit tests are included for all changed behaviours in:
tests/test_options.py, tests/collector/helpers/test_collector_runtime.py,
tests/collector/test_host_collector.py, tests/agent/test_host.py