feat: Stipple/Vue migration with isolated typed studies UI (squashed replacement for #6)

[hyperpolymath]

Squashed replacement for #6, which became undiffable after a history rewrite on the fork severed its merge base. Please review this instead of #6.

Why #6 stopped being reviewable

#6 was opened from hyperpolymath:main. That branch was then rewritten with git-filter-repo to strip ~260 MiB of dead blobs — committed node_modules (1,322 blobs), a long-gone web/ directory, and three raw FASTQ pools that no longer exist in the working tree:

	pack size	heaviest remaining path
before	269.35 MiB	data/MiSeq_SOP 170.80 MiB
after	9.23 MiB	data/MiSeq_SOP 7.71 MiB — the live .fastq.gz fixtures, the correct floor

~96.6% reclaimed, HEAD tree byte-identical at every stage. The six tracked data/MiSeq_SOP/*.fastq.gz fixtures are deliberately preserved.

The rewrite renamed every commit. GitHub resolves a merge base by SHA, so none of the commits #6 shared with this repository matched afterwards, and the base collapsed to the repository's Initial commit of 2026-01-30. Measured:

	before rewrite	after rewrite
merge base	recent shared commit	78845531 "Initial commit", 2026-01-30
ahead / behind	66 / 0	189 / 89
diff	313 files, +21,912 / −5,972	338 files, +65,566 / −22
mergeable	MERGEABLE	CONFLICTING
checks	4	0

+65,566 / −22 was not new work — it was the whole tree diffed from January. And a conflicting PR runs no pull_request workflows, so CI signal went to zero as well.

That state is not recoverable: the pre-rewrite head 81136ef7 still resolves through the API, but it is on no ref, so upload-pack refuses to serve it (not our ref).

What this PR is

One commit whose tree is byte-identical to hyperpolymath:main (tree 80269dd9), applied on top of this repository's current main (ecefb1c7).

• 319 files changed, +23,681 / −6,076 — reviewable as an ordinary change.
• Per-commit granularity from the original 66 commits is lost; the content is identical.
• Happy to split this into smaller reviewable parts if you would prefer — say the word.

Known state, stated plainly

• test/integration/test_server.jl:44 has been the CI blocker; the readiness loop was rebuilt as a wall-clock deadline with retry=false, because the original budget was measured at ~165 s rather than the ~30 s its comment claimed.
• SonarCloud reports C security / B reliability on new code. Per our standing rule a new scanner finding is an issue, not a merge blocker, but it is not hidden.
• frontend/ (React/TS) and ui/ (Genie/Stipple) both ship; docs/migration/STATUS.md lists what is not yet at parity.

🤖 Generated with Claude Code

https://claude.ai/code/session_01X3hgXxWm6umMgZkjYyHnnm

[JoshuaJewell]

The Genie direction is rejected and this tree still contains the hash(taxon_id) placeholder at Execution.jl:1423–1437. I would be happy to see the typed contracts/DTOs and the new work under frontend/tests/ in a future PR.