Add FakeJson and RebaseUrl transformers

[MustafaAldelaimi]

What

Two new transformers, motivated by the anonymised-local-dev epic's principle that local environments must run unmodified services — the strategy adapts to the product, never the reverse.

FakeJson — shape-preserving JSON anonymisation

Fakes a JSON value while preserving its shape: object keys, array lengths and value types are kept; string leaves become fake words of similar length; numbers, booleans and null pass through. Unparseable input collapses to {} (the EmptyJson behaviour) so nothing fails open.

Optional arg preserve_string_max_len: string leaves of at most N characters pass through unfaked, for enum-like option values ("Yes"/"No", option labels) that consuming services compare against. Free text (long strings) is always faked.

Why: strategies currently use EmptyJson on JSON columns whose fields downstream code reads (e.g. aurora Answers.answerBody.answerText, read by both the survey UI and the programme-recommendation scoring). Emptying the object crashes/short-circuits unmodified consumers; faking the leaves preserves consumability while stripping PII.

RebaseUrl — rebase stored URLs onto a local origin

Replaces a URL's scheme+host(+port) with a base arg, keeping path and query (path segments are Identity-preserved ids, so deep links keep resolving locally). Optional only_hosts arg (comma-separated): only URLs on a listed host are rebased; anything else — third-party links (survey tools, video links) or non-URL values — passes through unchanged. Without only_hosts, values with no recognisable origin become base alone, so a production host never survives. Optional template arg rebuilds the value from sibling columns of the same row (same mechanism as id_column), e.g. {base}/survey/skills-scan/{external_id}.

Why: URL columns are currently either scrambled (dead links) or passed through (links to production). Both break local journeys; per-service post-restore SQL rewrites are the current workaround. Real data mixes service URLs with third-party ones in a single column (measured in user_home user_tasks.action_url: ~84.6k multiverse-host rows vs ~1.2k qualtrics), hence only_hosts.

Notes

• Both transformers handle Postgres COPY text escapes (\\, \n, \t, \r): FakeJson unescapes before serde_json parsing and re-escapes on output.
• No new dependencies (serde_json and fake were already in the tree).
• 15 new unit tests; cargo test 188 passed, clippy and fmt clean.
• Functionally smoke-tested against real data: 221 real answerBody rows anonymised → all valid JSON, key/type/array-length distribution identical before/after, zero original strings survive.
• Strategy-side adopters queued in the epic: aurora Answers.answerBody/details → FakeJson; user_home user_skills_scans.url + user_tasks.action_url → RebaseUrl.

🤖 Generated with Claude Code