Resolve all four file-validation bugs from #135

[stevenweaver]

Summary

Addresses all four bugs from #135 — fixes the actual user-facing failures (Bugs 3 + 4) and the two telemetry gaps that surfaced them (Bugs 1 + 2).

Bug 1 — Unknown errors missing message field (+page.svelte)

338 of 342 `unknown` file-validation-error events had no `message` field in telemetry because `errorMsg = error.message || ''` was empty when something threw without a string. Replace the empty-string fallback with a defensive chain (`error.message → error.name → error.toString() → sentinel`) so the analytics payload always has something to work with.

Bug 2 — HTML responses parsed as JSON (WasmAnalysisRunner.js)

~290 `analysis-start-blocked` events surfaced raw V8 parse errors like `Unexpected token '<', '<!doctype'... is not valid JSON`. The actual culprit was `WasmAnalysisRunner.js:448` calling `safeParseJSON` without first checking whether `fetch(jsonBlob)` had returned HTML (which happens when HyPhy fails to produce a result file and the path resolves to a 404 page). Mirror the existing guard from `+page.svelte:805` so the user sees an actionable message instead of a parser error.

(The issue suggested `BackendAnalysisRunner.js` was at fault, but the backend uses socket.io — there's no `response.json()` there to guard. The real source was the WASM result fetch.)

Bug 3 — Neighbor-joining tree blocked without fallback (AnalyzeTab.svelte)

177 `analysis-start-blocked` events with `No neighbor-joining tree available. Please select a different tree source or upload a tree.` Root cause: `selectedTreeSource` defaults to `'inferred'`, but if datareader didn't produce an NJ tree (`FILE_INFO.nj` missing), the TreeSourceSelector disables the radio button while the underlying state stays `'inferred'`. Users then click Run and `getSelectedTreeData()` returns empty.

Add a reactive guard that switches `selectedTreeSource` to whichever source is actually available (`'uploaded'` if present, otherwise `'upload-new'`). Also tighten the error message when no fallback exists so users understand what to do.

Bug 4 — Genetic code parameter receiving invalid values (MethodSelector.svelte + WasmAnalysisRunner.js)

19+ `'X' is not a valid value for parameter 'Choose Genetic Code'` events with values like `Vertebrate`, `All`, `Source`, `0.1`. Root cause: aioli's `exec()` does a plain `command.split(' ')` with no quote awareness (confirmed by reading the aioli worker source). Multi-word genetic code names like `'Vertebrate mitochondrial'` get split into `--code Vertebrate` + a stray `mitochondrial`, which causes HyPhy to reject `Vertebrate` and then consume the next option's value as a fallback — explaining `'All'` (from `--branches All`), `'0.1'` (from `--pvalue 0.1`), and `'Source'` (from RELAX branch labels).

Pass the numeric `geneticCodeId` (always single-token) as `--code` instead of the descriptive name. The backend already does this via socket params.

Test plan

• All 201 existing tests pass
• Smoke test: trigger a HyPhy failure that produces no result file → confirm "HyPhy X analysis did not produce a result file" toast instead of the parser error
• Smoke test: throw something without a message string → confirm `file-validation-error` events include a `message` field
• Smoke test: upload an alignment where NJ inference fails → confirm `selectedTreeSource` auto-switches and the error message is helpful
• Smoke test: run a method with a multi-word genetic code (e.g. "Vertebrate mitochondrial") → confirm HyPhy now receives the numeric code and accepts it

Follow-ups (out of scope)

• `WasmAnalysisRunner.js` builds args as a space-joined string. The space-split issue affects more than just `--code` — any arg with a multi-word value (file paths, branch labels) is vulnerable. A larger refactor to pass `args` as an array to `aioli.exec(command, argsArray)` would eliminate the entire class of bug.
• Bug 3's deeper question is why the NJ tree fails to build for some inputs. The current fix routes around the problem; understanding the root cause (small alignments? non-standard characters? specific datatypes?) would be its own investigation.