[rahlk]

Closes #13.

This PR updates the install, guide, example and backend pages for cldk 2.0.0, the stable release on PyPI. The API reference pages under reference/python-api/ are in #14.

Each sample on these pages ran on cldk==2.0.0 from PyPI, and each output comment comes from such a run. The pinned samples are Apache Commons CLI 1.7.0, Flask 3.1.3 and Zod v4.6.7. The analyzers are codeanalyzer-python 1.5.3, codeanalyzer-typescript 1.6.0 and codeanalyzer-java 3.3.3, with JDK 17 and Maven 3.9. Each page names the versions that produced its outputs.

Merge order

Merge #15 first, then this PR, then #14. Merge this PR and #14 back to back: #14 links to these pages, and on astro alone these pages still describe the release candidate.

Changes

Install and start (installing, quickstart, resources/cheatsheet, what-is-cldk, examples, reference/sdks)

• Install 2.0.0 with no --pre. Every extra is quoted for pip, Poetry and uv, and every Java sample installs cldk[java].
• State the Python versions (3.11 to 3.14, a C compiler on 3.14, no 3.15 yet).
• SYSTEM_PYTHON for each kind of virtual environment, because CLDK.python() raises RuntimeError in one without it.
• State the JDK and Maven that Java call graphs need, and the JAVA_HOME rule.
• Use real outputs. Commons CLI has 75 classes. Its call graph has 652 nodes and 1980 edges on JDK 17, and 1976 edges on JDK 21 or 25.

Guides (concepts, symbol-tables, call-graphs, dataflow, system-dependence, taint)

• Remove the release-candidate badges and notes. Rerun the outputs on 2.0.0.
• cldk 2.0.0 pins codeanalyzer-python 1.5.3, which still has codeanalyzer-python#220. The pages keep the 1.5.4 workaround, and they state the gaps that 1.5.4 still has, for example sanitize(PREFIX).
• Correct what paths, exhausted, complete and unresolved claim.
• The upgrade notes give pip install -U "cldk[java]" for Java readers.
• Terminology: the guides say "level" for the four analysis levels, never "rung". The heading "The ladder CLDK builds" is now "The four levels CLDK builds", and backends/codeanalyzer-iac.mdx links to its new anchor.

Examples (examples/java, python, typescript, javascript)

• Clone Flask 3.1.3 and Zod v4.6.7 instead of HEAD, so the outputs stay valid.
• Regenerate every output. Most Java outputs did not match cldk 2.0.0.
• The JavaScript page gives every file of its my_app and mixed_app samples.

Backends, scale and contributing

• Update the CLI flags and output schemas to the pinned analyzers.
• State the cache rules that apply through the SDK.
• Quote "cldk[neo4j]".
• Remove the invented Odoo outputs.
• State the TypeScript Neo4j attach limit for declaration merging.

How the changes were checked

• Writers and reviewers. For each group of pages, a writer ran every block on a fresh copy of the pinned sample. A runtime reviewer ran every block again, and an editorial reviewer checked the text against the SDK source and the style rules. A fixer applied the defects, and a recheck confirmed the fixes. A final pass rechecked the last edits and compared these pages with the reference pages in Python API reference: correct and complete it for cldk 2.0.0 #14.
• Install paths. Each install line ran as pasted into zsh. Other runs used Python 3.9, 3.10, 3.14 and 3.15, uv and Poetry projects, and a Fedora container.
• Java toolchain. Runs covered JDK 11, 17, 21 and 25, a missing JDK or Maven, and a cache from cldk 1.5.0.
• Neo4j. The at-scale pages ran against a private Neo4j instance.
• Build and links. npm run build passes. 0 of 2104 internal links on the changed pages are broken.
• Merged state. A build of astro with ci(api-docs): build the reference from the latest stable SDK tag #15, this PR and Python API reference: correct and complete it for cldk 2.0.0 #14 merged passes the link validator from Python API reference: correct and complete it for cldk 2.0.0 #14 (4949 internal links, 0 broken).
• Text rules. The rendered prose has no em dash, no --pre, no release-candidate text and no "rung".

Known gaps

• at-scale/kubernetes.mdx keeps the Odoo walkthrough as a labelled template with no measured output.
• backends/codeanalyzer-python.mdx keeps my_pkg in its canpy command templates.
• A second git clone into an existing commons-cli, flask or zod directory fails. The pages do not say to remove the old directory first.
• The Flask call graph has 786 or 787 nodes between fresh runs, because Jedi resolves two read() calls in different ways. The page says so.

SDK problems that these pages work around

1. The SYSTEM_PYTHON workaround. codeanalyzer-python reads sys.base_executable, which CPython does not have, so a run from a virtual environment can raise RuntimeError. Present in 1.5.3 through 1.6.2.
2. Stale caches. Java and TypeScript reuse a cached analysis with no check of the sources. TypeScript also reuses a cache from a lower analysis level.
3. The 1.x-cache error suggests eager_analysis=True. That keyword raises TypeError. The keyword is eager=True.
4. CLDK.java(..., target_files=...) always fails on codeanalyzer-java 3.3.3.
5. The TypeScript Neo4j backend cannot attach to the Zod graph (declaration merging).
6. cldk 2.0.0 does not install on Python 3.15.