Add problem areas: Tekton pipeline review, migration path, multi-tenancy

[vdemeester]

What

Three new problem documents and practitioner perspective additions to the code review doc.

New problem areas

Tekton Pipeline Review — Reviewing Tekton task/pipeline definitions is a distinct discipline from reviewing application code. Build-definitions is the most critical repo in the org and it's almost entirely Tekton YAML with embedded shell. This doc covers:

• Embedded shell scripts in YAML (bash semantics + YAML embedding + Tekton parameter substitution)
• Stringly-typed interfaces (result propagation, parameter threading, workspace bindings)
• The trusted task model and its security implications for review
• Backwards compatibility and versioning concerns
• What a pipeline review agent needs to understand

Migration Path — How to get from today's workflow to agent-driven development incrementally. Proposes 5 phases:

• Phase 0: Observation (readiness assessment, baseline data)
• Phase 1: Shadow review (agents comment, humans still decide)
• Phase 2: Assisted review (agent review required, human merge)
• Phase 3: Conditional autonomy (Tier 0 auto-merge on graduated repos)
• Phase 4: Full autonomy for graduated repos

Includes repo selection criteria, bootstrap requirements, and anti-patterns.

Multi-tenancy — Konflux is multi-tenant. Agents need to understand tenant isolation boundaries to avoid introducing multi-tenancy bugs. Covers:

• The dual nature of multi-tenancy (workspace isolation + build-time isolation)
• Common multi-tenancy bug patterns (missing namespace scoping, cross-namespace references, shared state without isolation)
• The testing gap (test coverage numbers overstate readiness for multi-tenancy correctness)
• How multi-tenancy fits in the sub-agent model

Additions to existing docs

Code Review — Added two new sections:

• The heterogeneous codebase problem — the sub-agent model assumes Go code, but the org is heterogeneous (Go, Tekton YAML, shell, React, Python). Different content types need different review approaches.
• What human reviewers actually catch (and miss) — grounding the sub-agent model in real review experience. What agents will struggle with ("this works but it's the wrong approach"), what agents could be better at (consistent convention enforcement, cross-file impact analysis).

Updated README.md and CLAUDE.md to link the new docs.