docs: fill stream operator doc gaps and add missing overloads

[pjfanning]

Motivation

Some stream operator pages have empty sections or stubs, several pages
list only some of an operator's overloads, and the operator index
generator still lists removed operators as pending.

Modification

• Unzip.md, UnzipWith.md: fill the empty Signature sections
• Balance.md: add the missing Signature section
• Source/items.md: remove the empty Examples heading (no snippet exists)
• stream-flows-and-basics.md, stream-dynamic.md: remove empty
  Introduction headings
• stream-refs.md: replace the two-line Bulk Stream References stub with a
  "Transferring large amounts of data" subsection under Stream References.
  It explains using the existing SourceRef/SinkRef for large transfers
  (there is no separate bulk API) and lists the caveats: back-pressure
  across the network, single-use refs, no automatic resumption, the
  remoting frame size limit and the subscription timeout
• StreamOperatorsIndexGenerator.scala: drop the removed throttleEven,
  actorPublisher and actorSubscriber from the pending lists
• Add missing overloads and describe their extra parameters:
  dropRepeated(p), flatMapConcat(parallelism, f),
  groupedWeightedWithin(maxWeight, maxNumber, d)(costFn),
  takeWhile(p, inclusive), zipLatestWith(..., eagerComplete),
  Compression.gzip/deflate level and autoFlush variants, and the
  onErrorContinue predicate and Java variants

Also fix stream docs whose text contradicts the code (often text copied from a neighbouring operator):

• Guides: Unzip name and MergeSequence output (stream-graphs), pull() precondition (stream-customize), KillSwitch is a BidiFlow (stream-dynamic), Java failure wording (stream-error), Scala vs Java framing truncation (stream-io), NotUsed package (stream-quickstart), drop the "founding member" claim (stream-introduction, stream-flows-and-basics)
• Source pages: asSubscriber/fromPublisher signatures and links, iterate parameter names, lazySource, maybe (completionStage), queue completion
• Sink pages: queue backpressure, collect, completionStageSink, futureSink, lazyCompletionStageSink, lazyFutureSink (sink, not flow)
• Flow pages: flattenOptional description; remove the wrong second "completes" line from completionStageFlow, futureFlow, lazyFlow, lazyFutureFlow, lazyCompletionStageFlow and add missing "cancels" lines
• asJavaStream: elements come from upstream
• Source-or-Flow pages: alsoToAll attaches Sinks, batch has no costFn, buffer fail case, delay/delayWith take DelayOverflowStrategy, foldWhile completion, interleave/interleaveAll backpressure when downstream backpressures, limitWeighted accumulated cost, mergeLatest example and eagerComplete, onErrorContinue fails on unhandled errors, wireTap Sink variant signatures, zipLatestWith combine function, Source.sliding signature

Every signature was checked against the scaladsl/javadsl sources. The new @apidoc anchors are first checked by the CI docs build. This PR merges cleanly with the other open docs PRs (#3583, #3587, #3588 and others; checked with git merge-tree).

Adding reference pages for operators that have none (Sink.foldAsync, Sink.contramap, Flow.fromFunction, Flow.join, the processor methods and the for-comprehension support) is tracked in #3589.

Result

Operator pages list all overloads, have no empty sections, and describe what the code does.

Tests

• Not run - docs only (plus removing stale names from the docs index
  generator's pending lists)

References

Refs #3589

[He-Pin]

I verified all 27 documented overload entries against the real scaladsl and javadsl sources and they all match, including the groupedWeightedWithin argument order and the @since 1.2.0 tag - no mismatches, and the pendingSourceOrFlow deletions are genuinely dead (throttleEven removal is confirmed by the 2.0.x MiMa excludes). Two things before this can go in: (1) Check / Tests (scala3) is red here, so it cannot be merged until that job is re-run and green - I do not want to assume it is just the 6h timeout; (2) the iterate.md page below is only half aligned. Note also that dropRepeated.md:8 still has the malformed scala="#dropRepeated):FlowOps..." anchor, which I believe #3587 already fixes - worth a look so the two do not conflict.

[He-Pin]

The signature fix here is right - the real parameters are p and f (Source.scala:559 def iterate[T](seed: T)(p: T => Boolean, f: T => T)) - but this sentence you just edited now names next, a parameter the new signature says does not exist. The page ends up using four names for two parameters: the Description above still says hasNext predicate and next function, and the Reactive Streams semantics below still say `next` function returns and `haxNext` predicate (the latter is also misspelled). Could you align the whole page to p/f while you are in here?

[He-Pin]

Pushed b06f22e aligning the whole iterate.md page with the real parameter names p/f (Description, warning, RS semantics), fixing the haxNext typo and completing the unfinished "The next example shows how to craet" sentence. The previous Tests (scala3) run was cancelled at the 6h timeout; this push re-triggers CI - will merge once it is green.

[He-Pin]

Merged current main into the branch to clear the conflict with #3588 (it had meanwhile fixed haxNext -> hasNext and the unfinished Examples sentence on main). Resolution: kept main's more precise "counts from 1 to n" example sentence, kept this PR's p/f naming everywhere (matches Source.scala:559), and #3587's dropRepeated.md anchor fix survived the auto-merge. CI is re-running; will merge when green.

[He-Pin]

Rebased onto current main (8a012b9) instead of the merge commit. Only iterate.md conflicted (with #3588's meanwhile-merged fixes); resolution keeps main's "counts from 1 to n" example sentence and this PR's p/f naming everywhere else. Branch tip is now 620727c; CI re-running.

[He-Pin]

lgtm