docs: replace unresolvable private tracker references in source and test comments

[TKaltofen]

Summary

Source and test comments cite issue identifiers from a private tracker that resolve to nothing in this repository, so a reader cannot follow them.

Motivation

Twelve files carry comments and docstrings that justify a design decision by pointing at an identifier no public reader can look up. Four occurrences are in shipped source, the rest in tests. Two examples of what this costs:

• A comment explains that a raise "now surfaces as a framework defect" and attributes the change to an unresolvable identifier, so the reasoning is unverifiable.
• A test module docstring states its contract is defined by a design document that is not in the repository.

Until recently memory-bank/ gave these identifiers at least partial context. #889 removes it, so nothing in the repository defines them any more.

Where a public issue covers the same change, the fix is to cite that instead: several of these sites already pair the private identifier with a public number (#763, #772, #798), which shows the public reference alone would have been sufficient.

Code pointers

Shipped source (4 occurrences):

• mloda/core/abstract_plugins/components/feature_chainer/feature_chain_parser.py:58, :527, :873
• mloda/core/abstract_plugins/components/feature_chainer/feature_chain_parser_mixin.py:307

Tests (10 files):

• tests/test_core/test_prepare/identify_seam.py, test_identify_seam.py, test_resolve_or_raise.py, test_candidate_elimination_reasons.py, test_first_pass_rejection_recording.py, test_abstract_feature_group_not_instantiated.py, test_compute_framework_capability.py
• tests/test_core/test_abstract_plugins/test_components/feature_chainer/test_option_value_rejection_never_escapes.py, test_retire_transitional_seams.py
• tests/test_plugins/test_undeclared_option_reads.py

Find them with grep -rn "os-0[0-9][0-9]" mloda/ tests/.

Definition of done

• Every such reference is replaced by the public issue number that covers the same change, or by a self-contained explanation, or deleted where it adds nothing
• No reference to a repository-external design document survives without the contract being stated in the test or the docs
• grep -rn "os-0[0-9][0-9]" mloda/ mloda_plugins/ tests/ returns nothing
• tox passes

[TKaltofen]

Closing as not worth a standalone change. These are comments and docstrings: they affect no behavior, no test, and no user of the library, and each one still reads as "a review prompted this" even when the identifier itself resolves to nothing. Weighed against churn on files that several PRs touch concurrently, a dedicated sweep is not a good trade.

The better fix is opportunistic: when a change already edits one of these files, rewrite the reference it touches into self-describing prose, or point it at the public issue where one exists. Reopen if the references start actually misleading someone.