Fix bulk add writing wrong values and dropping rows and columns

[labkey-martyp]

Rationale

Fixes a set of bugs that let the EHR bulk add importer write wrong or incomplete data without reporting anything, in the UI, the browser console, or the server log.

Because nothing fails, the damage is only discoverable by diffing the loaded data back against the source, so an import can appear clean while carrying shifted dates, missing dates, lookup values swapped for unrelated ones, whole columns absent because their header was spelled differently, and — on tables whose key is assigned by the server — only the last of the pasted rows. The reporting added here echoes pasted text back to the user, so it is escaped; an import commonly starts from a spreadsheet someone else produced.

The same work also covers the opposite failure, where the importer refused input that was perfectly good or blamed the wrong thing for it, since a rejection nobody can act on costs the user as much as a silent one.

Related Pull Requests

None.

Changes

• Every date column is now parsed the same way as the form's primary date field. Previously the others were read as UTC, so in a negative-offset timezone they landed a day early, and a date falling on the epoch was discarded as empty.
• Lookup and project values now require an exact match rather than resolving on a leading substring, so an input can no longer be replaced by an unrelated record that merely starts with the same text. Matching stays case-insensitive, ignores surrounding whitespace, and accepts either the display value or the stored key.
• A value that cannot be resolved to a lookup, a project, or a date is now reported against its row rather than being written through as raw text or left silently empty, as is a lookup whose values have not loaded. The import stops so the source can be corrected first.
• Pasted text is escaped wherever it appears in those error messages.
• Column headers are now identified the same way the rest of the product identifies a pasted column, by a field's name, label, or any of its import aliases, ignoring case and surrounding whitespace. Previously only the exact field name or a single alias was tried, so a header spelled any other way dropped its whole column without a word.
• A header naming no field in the form, or two headers naming the same one, is now reported and the import stops before any row is read. Note the behavior change: a spreadsheet carrying extra columns the form does not have will no longer import until those are removed, where previously they were ignored.
• Trimming the pasted block no longer strips tabs, only spaces and line endings. Stripping tabs discarded the empty trailing cells of the final row, which then looked truncated and was rejected even though every value in it was correct.
• A row is measured against the last column a required field occupies rather than against the count of required fields, which bore no relation to where those fields sit. A row missing only optional trailing cells is no longer rejected for it.
• A problem belonging to the header row — a required column absent, or a header row naming nothing at all — is reported once instead of restated as a missing value on every one of up to 250 rows.
• Pasting several rows into a table keyed on a server-assigned column no longer collapses them onto one record, which previously left only the last row's values.