docs: explain how to parallelize work within a single stream

[dbrattli]

Follow-up to #713.

Problem

The "CPU Concurrency" section of the getting-started guide demonstrates concurrency across three independent pipelines, but never states that the items of a single pipeline are delivered one at a time. Readers reasonably conclude that adding observe_on(ThreadPoolScheduler(n)) will process items in parallel, then report the resulting sequential execution as a scheduler bug.

This is by design — ScheduledObserver drains one notification at a time to honour the Rx contract that on_next never overlaps — but nothing in the docs says so. It has come up repeatedly: #713, #67, #100, #528.

Changes

• docs/get_started.rst — new "Parallelizing Work Within a Single Stream" section showing the sequential-vs-parallel comparison and the flat_map fan-out pattern (both the just(...) + subscribe_on form and the more direct from_callable(..., scheduler=...) form), plus a note on the concat_map ordering trade-off and merge(max_concurrent=...).
• docs/get_started.rst — an .. important:: admonition in the existing subscribe_on/observe_on discussion, linking to the new section.
• reactivex/operators/_observeon.py — matching note in the observe_on docstring, so the clarification is reachable from the API reference and from IDE tooltips.

No library behaviour changes.

Verification

Every snippet was executed against this branch. With a 2-second intense_calculation over three items:

	wall clock	order
observe_on(pool) + map	6.0s	preserved
flat_map + subscribe_on(pool)	2.0s	completion order
flat_map + from_callable(..., scheduler=pool)	2.0s	completion order
concat_map + from_callable(..., scheduler=pool)	6.0s	preserved

which matches what the new section claims, including the ordering caveat.

sphinx-build -W --keep-going produces 67 warnings on this branch and 67 on master — no new ones. ruff check, ruff format --check and pyright pass clean on the changed module, and tests/test_observable/test_observeon.py passes.

🤖 Generated with Claude Code

[coveralls]

coverage: 93.495%. remained the same — docs/parallelism-within-stream into master