N TypeInfo reads the wrong token and switches on token kind

• Track N (compiler/pyparser.inc, the AN_TYPEINFO parse around line 42435).
• Sibling of the Pascal-side fix in compiler/pasparser_expr.inc (feature-typeinfo-all-types, landed on Track A). Filed rather than fixed: pyparser.inc is Track N's file and N work is currently deferred, so this is a hand-off, not a half-applied change.

The two defects, both visible in one 17-line block

tiName := GetTokenStr(TokPos);      { (1) off by one }
case CurTok.Kind of                 { (2) resolves by KIND, not spelling }
  tkInteger_T:  tiTk := tyInteger;
  tkBoolean_T:  tiTk := tyBoolean;
  ...

(1) off by one. lexer.inc's Next reads Tokens[TokPos] and then does Inc(TokPos), so by the time this line runs TokPos already points at the ) after the type name. tiName is therefore ), not the type. It has been invisible because tiName feeds only the 'TypeInfo is not supported for type: ' error, and every branch that reaches that error had already failed to assign a kind — so the wrong string only ever showed up inside a message nobody diffed. The Pascal copy now reads GetTokenStr(TokPos - 1).

(2) kind, not spelling. byte and integer both lex as tkInteger_T (two spellings, one token kind — see lexer.inc), as do the other width-variant ordinals. Switching on CurTok.Kind therefore collapses them: TypeInfo(byte) answers Integer. The Pascal copy now tries OrdinalNameToTk(tiName) first and falls back to the kind switch only when the spelling does not resolve.

Why it is a real N bug, not a divergence

NilPy is upward compatible with CPython, but this is not a CPython-parity question at all — it is a wrong answer from a construct NilPy accepts and compiles. A program that branches on the reported type takes the wrong branch.

Fix

Port both halves from compiler/pasparser_expr.inc (search tiName): TokPos - 1, then OrdinalNameToTk before the case. Keep the explanatory comments — the off-by-one is the kind of thing that gets "cleaned up" back into existence.

Gate

make test-nilpy green + self-host byte-identical, per Track N's lane rules. A .npy assertion in the same shape as test/test_typeinfo_scalar_names.pas (assert the byte-vs-integer pair, since that is the case the kind switch cannot express) is the test that bites.