Clearer parse error for identifiers with a '-' in the middle (#7742)

[Unisay]

Summary

Closes #7742.

A UPLC file emitted by Scalus 0.16.0's toUplcOptimized — the HTLC sample from the issue — fails to parse with a confusing message hundreds of lines from the real site. The shape reduces to one bad identifier:

(program 1.1.0 (lam pubKeyHash-305478r71 (lam xIn72 xIn72)))

- isn't an identifier character in UPLC (see isIdentifierChar in PlutusCore.Name.Unique); it exists only as the separator between the name text and its numeric unique-suffix (foo-123 → Name "foo" (Unique 123)). So pubKeyHash-305478r71 isn't a valid unquoted identifier. The old parser still "succeeded" on it in the worst possible way: ate pubKeyHash as the name, -305478 as the unique, left r71 in the stream. r71 then got picked up as the body of the enclosing (lam ...), leaving the closing-paren bookkeeping off by one, and megaparsec eventually blew up on an unrelated ( with "unexpected '(' expecting ')'" at line 448 col 39 — nowhere near the bad name on line 447.

This PR keeps the grammar as-is (it should stay narrow — see the discussion in the issue comments) but makes the diagnostic point at the real problem.

Before / after on the HTLC file

before: htlc.uplc:448:39: unexpected '(' expecting ')'
                            — on a lambda 8+ characters past the real site,
                              on the wrong line

after:  htlc.uplc:447:41: Invalid identifier 'pubKeyHash-305478r71'
        A '-' inside a name is the numeric unique-suffix separator and must be
        followed only by digits and a word boundary.
        To use this text as a name verbatim, quote it with backticks:
        `pubKeyHash-305478r71`
                            — on the offending name, with an actionable hint

[Unisay]

The single Hydra failure (ci/hydra-build:x86_64-linux.ghc96-mingsW64:checks:plutus-core:test:flat-test) is unrelated to this change. Log says:

stable byte encodings: FAIL
Exception: diff: startProcess: does not exist (No such file or directory)

It's the Windows-cross builder missing the diff binary that tasty-golden shells out to. The same check failed on the recently-merged #7734 with the same message. All native x86_64-linux / aarch64-darwin / x86_64-darwin Hydra checks pass, and locally ./scripts/regen-goldens.sh (~12 800 tests, 7 subprojects) produces zero golden updates outside the three this PR adds.