Allow backtick-quoted names in expressions

[hadley]

A table or column could be declared with any non-empty name, but only a plain identifier ([A-Za-z_][A-Za-z0-9_]*) could be referenced from an assertion or a join — so a column called creation date was declarable and then unreachable, with no quoting syntax to bridge the gap.

Names of any other shape — containing a space or punctuation, starting with a digit, or colliding with a reserved word — now go in backticks:

constraints:
  - assert: '`creation date` <= `end`'
relationships:
  - join: '`other studies`.`study id` = study.id'

A quoted name may hold any character, doubling a backtick to embed one. Each side of a join's . is quoted on its own, so a . between backticks is part of the name. Quoting changes only how a name is read, never how it's matched, so backticks are optional on a name that doesn't need them. Malformed quoting (unterminated, empty) lands on the existing S19/S04 rather than a new check code, so validation.md and schema.yaml are unchanged.

Backticks rather than SQL's " because '…' is already a string literal (and COLUMNS('regex') a regex), and because it matches the analytics mini-languages users come from — pandas query(), dplyr, Spark. One wrinkle documented in the spec: a backtick can't open a plain YAML scalar, so an expression starting with one needs YAML quotes; one merely containing a backtick doesn't.

Spec first: site/expressions.md gains a Column references section, plus grammar and COLUMNS([...]) updates.

Tests: quoted names, reserved-word names, embedded backtick/dot/non-ASCII, spans, and the unterminated/empty rejections in both parsers; tests/fixtures/spec/quoted-names-ok.yaml exercises the whole path through the CLI.

Fixes #148, #137

🤖 Generated with Claude Code

[gaborcsardi]

now go in backticks:

There are no backticks in this example. :)

[gaborcsardi]

Maybe we should de-duplicate the two parsers? E.g. this PR adds parse_quoted_name() to both, and the two implementations are exactly identical. Which makes sense, they are supposed to be identical. But then we should probably merge the two.