feat: add the corrective fact to CLI failure diagnostics

[omer-cengel]

Summary

sqlcj generate already reported each failure as a single sqlcj: line naming the failing file, but many messages stopped there, and SQL syntax errors either said only Failed to parse schema or leaked a parser exception class name. Every configuration, source, SQL syntax, and output diagnostic now keeps its existing prefix and ends with the fact that explains the failure, so the message says what to fix.

Changes

• Unreadable configuration files and schema or queries sources end with the filesystem reason, such as No such file or directory or Is a directory.
• A malformed configuration file reports the YAML parser's problem with its one-based line and column.
• Failures to create an output directory, write a generated file, read or write the output manifest, or delete a stale generated file end with the filesystem reason.
• A schema syntax error reports the unexpected token with its line and column; a query syntax error reports the parser's reason after the query's existing header location.
• Lexical SQL errors, such as an unterminated string literal, use the parser's own wording; no diagnostic names a parser exception class.
• The README, configuration guide, and PostgreSQL guide state exit status 1 for every failure category and show the new messages.

Scope and non-goals

• Message prefixes, the single-line format, and exit status 1 are unchanged; only the trailing fact is new.
• No CLI options, runtime changes, or generated-code changes are included.
• Runtime-version mismatch is not detected by the CLI; the documented rule that the CLI and runtime are used at the same version still covers it.