[zaleslaw]

Closes #1975

The public merge APIs had no KDoc at all. This PR adds concise KDoc for all 12 non-deprecated
merge APIs, with every example backed by a runnable test, and rebuilds the merge page of the
documentation website to match. Two bugs found while writing the tests are fixed.

Behaviour change

notNull() now applies in intoList() too: df.merge { a and b }.notNull().by(" ").intoList() skips
null values (before: returned them all). into(path) now creates missing parent column groups:
into(pathOf("person", "fullName")) merges into a new nested group instead of throwing
Column 'fullName' not found among [top-level columns]. Single-segment paths, replacement of existing
targets and the full core suite are unaffected.

Files

File	What changed
core/src/main/kotlin/org/jetbrains/kotlinx/dataframe/api/merge.kt	KDoc for the 12 non-deprecated APIs (MergeDocs grammar, shared KoDEx snippets); notNull filter shared by into/intoList; into(path) group selection made path-aware
core/src/main/kotlin/org/jetbrains/kotlinx/dataframe/documentation/DocumentationUrls.kt	new Merge website-link entry
samples/src/test/kotlin/org/jetbrains/kotlinx/dataframe/samples/api/MergeSamples.kt	12 tests backing every @sample in the new KDocs and every sample of the rebuilt page, with Sample Helper outputs
core/src/test/kotlin/org/jetbrains/kotlinx/dataframe/api/merge.kt	17 behaviour and claim tests
docs/StardustDocs/topics/merge.md	page rebuilt: Writerside summaries, corrected grammar (transform, not merger), full notes incl. notNull/asStrings/into(path), Parameters section, split round-trip example
docs/StardustDocs/topics/_shadow_resources.md, docs/StardustDocs/resources/api/merge/	10 new resource entries and 10 generated HTML outputs of the page samples (first HTML this page ever had)
samples/build.gradle.kts	merge.md registered in the Korro docs config
core/src/test/kotlin/org/jetbrains/kotlinx/dataframe/samples/api/Modify.kt	the five legacy merge sample functions deleted — page samples migrated to :samples

Problem 1: notNull() silently ignored in intoList()

• Problem: df.merge { a and b }.notNull().intoList() returned the unfiltered values, and an
  all-null column-group row stayed, while into("col") filtered both.
• Options: filter inside intoList; keep unfiltered and document the exception. Rejected the
  second: notNull is the documented contract of the merge pipeline, not an into-only option.
• Chosen: one shared private filterNotNullAndEmptyRows used by into and both intoList
  implementations. For callers: notNull now affects intoList results.
• Test: intoList excludes null values when notNull is set,
  intoList excludes an all-null row value from a column group when notNull is set (both failed before).

Problem 2: into(path) threw for any new nested path

• Problem: df.merge { age }.by{}.into(pathOf("person", "fullName")) threw
  Column 'fullName' not found among [top-level columns] — also for a new sub-path inside an existing
  group. Only single-segment paths and fully existing nested targets worked.
• Options: fix the selector inside into(path); pre-create parent groups in under. Rejected the
  second: the failure is in the convert selector — getColumnGroup(path) returns an eagerly resolved,
  path-less ColumnGroup that re-resolves by its leaf name only.
• Chosen: convert { columnGroup(mergePath) } — a path-aware accessor. For callers: missing
  intermediate groups are now created instead of an exception.
• Test: into a new nested path creates the missing column groups,
  into a new sub-path inside an existing column group (both threw before).

Tests

Before this PR, merge had no KDoc-backed examples; behaviour was covered only by a few
testSets/person2 tests. All expected values come from real runs.

Test	What it pins down
selection order; merged-column position; the String overload	values follow the selector order; the merged column takes the first selected column's place; merge("a", "b") equals the DSL form
by(separator, ...) join; null rendering	all 5 parameters reach joinToString; a null value renders as null
notNull on every final step	nulls and all-null group rows are dropped in into and in intoList, with the default and with a custom transform
by { } contract; asStrings	the transform receives the source row as receiver and the collected values as a list; asStrings() equals by(", ")
into target handling	existing targets are replaced entirely in place; missing parent groups are created
MergeSamples (12)	every @sample in the KDocs and every sample of the website page runs and produces the asserted result

Note for the reviewer

• The page is rebuilt beyond the KDoc the ticket asked for: it had no section the new KDoc links
  could point to, never explained notNull or asStrings, called the lambda merger where the API
  says transform, and had no generated HTML outputs.
• getColumnGroup(path) returns an eagerly resolved, path-less ColumnGroup; used as a column set
  inside select/convert selectors it re-resolves by its leaf name only and breaks nested paths —
  the path-aware columnGroup(path) is the correct form there. This was the root cause of Problem 2.
• DataRow.isEmpty() is used deliberately in the shared filter with @Suppress("DEPRECATION"):
  its prescribed replacement values().all { it == null } is not equivalent for row values collected
  from column groups — a probe showed the filtering invert there (DataRowImpl.values() is df-based,
  isEmpty() is owner-based). Should the ReplaceWith itself be fixed? Not changed in this PR.

[jetbrains-air]

Looks good to me — approving! 🎉

Produced by Air Automations. Name: AI Review / Run: https://air.jetbrains.cloud/org/05cf1a7f-6ab5-713b-abd3-29d0c8a05e2d/automations/8daf2412-0b6c-4433-90c0-b40ab5aebcdb?run=0ed70356-ce85-4d90-b3ca-e32f344e01c1