Support scientific-notation float literals (1.5e10, 1e6)

[CoreyRDean]

Summary (non-technical)

You can now write floating-point numbers in exponent / scientific notation — 1.5e10, 2.0e-3, 1e6 — and they evaluate to the value you'd expect. Before this change, writing 1.0e30 didn't just fail; it failed with a baffling Function 'e30' not found, sending you hunting for a typo in a function that doesn't exist. A real downstream consumer (rcce2) hit exactly this on Local huge# = 1.0e30 and had to rewrite every constant as a plain decimal and record the gotcha. Exponent literals are table stakes for a language; this closes the gap and makes more legacy/standard BASIC drop-in compatible.

Technical summary

The value pipeline was already exponent-correct: FloatConstNode converts its token via atof() (parser.cpp:1106, exprnode.cpp:385), and atof("1.5e10") / atof("1e30") parse correctly. The entire gap was in tokenization — toker.cpp's numeric paths scanned only mantissa digits, so 1.5e10 lexed as FLOATCONST(1.5) + IDENT(e10) and the parser read e10 as a call.

• src/blitzrc/compiler/toker.cpp: add a small scanExp helper that consumes an optional [eE][+-]?<digits> suffix, applied to all three numeric paths (.-leading, digit-with-dot, digit-no-dot). A bare-integer mantissa with an exponent is promoted to FLOATCONST (1e6 is a float, not an int).
• Conservative by construction: the e/E is consumed only when a full [eE][+-]?<digit> follows. 1.0e, 1.0eq, 1.0e+x keep their current behavior (1.0 + a separate identifier). No legacy program can depend on the old tokenization, because every newly-accepted form was previously a hard compile error.
• No parser/AST/codegen/runtime change. macOS unaffected (front-end only).

No breaking changes.

Acceptance criteria & results

• 1.5e3=1500, 2.0e-3=0.002, 1.0e+2=100, 1e6=1000000, .5e2=50 — pinned by Assert(FloatClose(...)) in tests/MathsTest.bb; blitzcc -t tests/MathsTest.bb passes (all 5 new blocks).
• Conservative guard: 1.5eq → Function 'eq' not found, 1.5e+q → Function 'e' not found (the e is not swallowed); 1.5e10 compiles. Verified at the CLI and pinned in a Test block.
• Formerly-failing Local huge# = 1.0e30 now compiles (exit 0).
• Full test.bat suite green ("Tests passed").
• Corpus sweep: 352 files, 325 compiled, 27 known-fail, 0 new-fail.
• lang_ref_basicdatatypes.html documents exponent notation with examples.

Trade-offs / deferred

• Promoted the no-dot form 1e30 to float deliberately (the common case; leaving it out would be a second sharp edge). Safe: an identifier can't start with a digit, so 1e30 was always INTCONST(1)+IDENT(e30) = a parse error before.
• Out of scope: hex/binary exponents, float suffixes, the Release/Recast/Reference contextual-keyword work (still the top legacy-compat backlog item).

🤖 Generated with Claude Code