[GreedyC]

Fixes #80.

Column documentation and generic tests using SQL double-quoted names in the manifest
now match the unquoted names in the catalog. For example, "Top Performer" matches
Top Performer; doubled embedded quotes are decoded. Manifest columns and test
references use the same normalization, retaining the existing case-insensitive
matching and leaving unquoted interior quotes intact.

Adds artifact-loading regressions for sources, models, seeds and snapshots, including
all three supported test-column reference locations, escaped quotes, ordinary mixed
case names and one-sided quote characters. README documents the matching semantics.

Validation:

• Before the production change: 24 failed / 64 passed; failures were the quoted-name
  cases. After the change: 88 passed.
• The complete unit suite passed on Python 3.10, 3.11, 3.12, 3.13 and 3.14 (88 each).
• The repository's tox unit environment passed (88 tests).
• The actual CLI computed 100% coverage from generated catalog/manifest JSON for
  quoted documentation, quoted tests, escaped-quote tests and ordinary mixed-case
  tests; all four output JSON reports had one hit and one covered column.
• pre-commit run --all-files passed all applicable hooks, including pinned isort and Black.
• Full tox was attempted: its unit environment passed, but the dbt integration dependency installation stalled and was interrupted. The integration environments therefore did not complete; no native warehouse/dbt integration success is claimed.

Limitations: JSON artifacts are synthetic representations of the reported artifact
shapes, not newly generated by a warehouse/dbt run. This change retains the existing
lowercase behavior; it does not add warehouse-specific case-sensitive matching or
other SQL quoting styles. The entire dbt 1.1–1.12 integration matrix is not claimed
as passing. Open PR #64 addresses missing table paths/generic test nodes and does not
overlap these normalization hunks. Issue #80 was open, unassigned, and had no claim
comments when checked; no external claim or PR was posted.

AI disclosure: implementation and tests were prepared with OpenAI Codex. The actual
GitHub Copilot CLI independently inspected the final source, regression and README
files in read-only mode and reported no actionable defects (session
551ac1f0-593c-4eae-ada5-ac3fe2a039a9). Copilot did not run tests or validate against
a warehouse. Review is advisory and does not represent maintainer approval.