refactor(nemesis): Remove Class:N count syntax from nemesis_class_name, enforce explicit list format

[cezarmoise]

Parallel nemesis runs using the space-separated nemesis_class_name format (e.g. 'SisyphusMonkey:1 SisyphusMonkey:1') were silently broken after the pydantic config migration (aa9cd75).

See https://argus.scylladb.com/tests/scylla-cluster-tests/1b6ed4e9-2c22-4153-94c4-3081d46c28d1/nemesis
Only 1 nemesis thread ran.

Summary

Replaces the implicit count/space-separated syntax for running parallel nemesis threads with an
explicit YAML list. Adds validation so misconfigurations are caught at config-load time rather
than producing silent surprises at runtime.

Problem

The old nemesis_class_name syntax accepted three formats simultaneously:

# Single string with count — one entry expanded to N threads
nemesis_class_name: "SisyphusMonkey:2"

# Space-separated string — split into multiple threads
nemesis_class_name: "SisyphusMonkey SisyphusMonkey"

# Explicit list (already worked, never formally documented)
nemesis_class_name: ["SisyphusMonkey", "SisyphusMonkey"]

This made nemesis_selector and nemesis_seed assignment opaque: if you had two threads and
one selector, was the selector applied to thread 0 only, or broadcast? The code used
selectors[i % len(class_names)], which silently cycled through selectors in ways that were
hard to reason about.

Changes

sdcm/tester.py — get_nemesis_class()

• The Class:N count expansion loop is removed.
• Explicit ValueError is raised if any entry in nemesis_class_name contains : (count
  syntax) or a space (space-separated format), with a clear migration message.
• Selector/seed assignment semantics are simplified and made explicit:
  • empty / not set → empty string / None per thread
  • single-element list → broadcast to all threads
  • list of length N (matching nemesis_class_name) → 1:1 mapping, enforced by config validator

sdcm/sct_config.py

• nemesis_class_name field description updated to document the new list-only API and
  explicitly deprecate the old syntax.
• New _validate_nemesis_parallel_config() method validates that multi-element
  nemesis_selector and nemesis_seed lists match the number of entries in
  nemesis_class_name. Mismatches raise ValueError with a helpful error message at
  config-load time. Called unconditionally from verify_configuration().

sdcm/nemesis/__init__.py — NemesisRunner.nemesis_selector

• __init__ parameter typed as Optional[str] = None.
• Property uses None as the sentinel for "not explicitly set" (replaces the old truthiness
  check that treated "" identically to None).
  • None → falls back to cluster.params.get("nemesis_selector"), enabling direct
    instantiation paths such as sct.py nemesis-list to pick up the config value.
  • "" → explicitly means "run all nemeses, no filtering".
  • Any non-empty string → used as-is.
• Config fallback handles both list (takes [0]) and str/None return types from params.

sdcm/utils/operations_thread.py

• Removed the isinstance(nemesis_seed, str) branch (the IntOrList type already resolves
  seeds to int or list[int]; string seeds no longer reach this code).
• Fixed the Random() seed guard from truthiness to is not None so that seed value 0 is
  not silently discarded.

YAML test-case files converted

All occurrences of Class:N and space-separated nemesis_class_name are replaced with
explicit YAML lists. Two files also had nemesis_seed values in the old space-separated
string format, converted to proper YAML lists/integers.

Migration guide

Before:

nemesis_class_name: "SisyphusMonkey:2"
# or
nemesis_class_name: "SisyphusMonkey SisyphusMonkey"

After:

nemesis_class_name: ["SisyphusMonkey", "SisyphusMonkey"]
# yaml list syntax also works
nemesis_class_name:
  - SisyphusMonkey
  - SisyphusMonkey

If you use nemesis_selector or nemesis_seed with multiple threads, the valid forms are:

# Broadcast the same selector to all threads
nemesis_selector: "not disruptive"

# Or map one selector per thread (must match nemesis_class_name length exactly)
nemesis_selector: ["not disruptive", "topology_changes"]

Any mismatch in list lengths will raise a ValueError at config-load time with a
descriptive message showing both lists.

---

Testing

• before: https://argus.scylladb.com/tests/scylla-cluster-tests/1b6ed4e9-2c22-4153-94c4-3081d46c28d1/nemesis
• after: https://argus.scylladb.com/tests/scylla-cluster-tests/46e3f41a-ebe8-4362-9664-03dc80457564/nemesis

PR pre-checks (self review)

• I added the relevant backport labels
• I didn't leave commented-out/debugging code

Reminders

• Add New configuration option and document them (in sdcm/sct_config.py)
• Add unit tests to cover my changes (under unit-test/ folder)
• Update the Readme/doc folder relevant to this change (if needed)

[cezarmoise]

@claude review this change please

[scylladb-promoter]

✅ Test Summary: PASSED

✅ Precommit: PASSED

Total	Passed	Failed	Skipped
25	15	0	10

✅ Tests: PASSED

Total	Passed	Failed	Errors	Skipped
1713	1698	0	0	15

Full build log

[cezarmoise]

Temporarily marked as draft as I want to improve further.

[cezarmoise]

This was wrong, confirmed by @timtimb0t