Fix #10: double-quote reserved-word field names in read_json_auto STRUCT type

[piotrszul]

Fixes #10.

Problem

pathsToSchema (src/ddb-sql-builder.js) emitted FHIR field names unquoted in the read_json_auto columns STRUCT type string. When a field name is a DuckDB reserved word — e.g. end from the FHIR Period type — the generated spec is unparseable:

columns = { period: 'STRUCT(start JSON, end JSON)' }
-> Invalid Input Error: Value "STRUCT(start JSON, end JSON)" can not be converted to a DuckDB Type.

Fix

Always double-quote the identifier in both emitted forms (root key and struct member):

return isInRoot ? `"${node.value}": '${sqlType}'` : `"${node.value}" ${sqlType}`

Always-quoting is safe for all names and avoids maintaining a reserved-word list (the issue's recommended direction). Verified in DuckDB 1.5.2 that quoting normal identifiers is harmless, that quoting works for both the root key and struct members, and that the previously-failing Encounter period.start/period.end view now executes and returns rows instead of aborting.

Tests

• Added a reserved-word (end) regression test in tests/schema.test.js.
• Updated existing schema expectations to the quoted output.
• All 192 tests pass (bun test), including the e2e tests that run real read_json_auto queries.

Note: the secondary schema-coverage observation in #10 (Period fields falling back to JSON instead of VARCHAR) is the R4-schema bug fixed separately in #11 — the two compose cleanly. This PR is purely the quoting fix.