fix(parser): preserve the original text of COND equations

[ThVerg]

Conditional expressions are rewritten into something no other tool accepts

Condition text is currently passed through Python's float(), and Verilog scalar constants
are split in half by the lexer:

condition in the source	before	after this PR
B==1'b0	B == 1.0 'b0	B == 1'b0
TE == 0	TE == 0.0	TE == 0
A == -1	A == -1.0	A == -1
A==+2	A == + 2.0	A == +2
B==1'b0&&C==1'b1	B == 1.0 'b0 && C == 1.0 'b1	B == 1'b0 && C == 1'b1
A==1.5	A == 1.5	A == 1.5 (unchanged)
~EN	~ EN	~ EN (unchanged)

Cause

Two independent mechanisms.

Scalar constants split. FLOAT and SCALARCONSTANT overlap: the constant's leading
digit is optional ([01]?), so for 1'b0 the lexer matches FLOAT on the 1 first and
SCALARCONSTANT on the remaining 'b0. equation() then rejoins the pieces with a space.

Numbers lose their text. equation shares the FLOAT terminal with delay values, and
the transformer's FLOAT() callback converts every FLOAT token to a Python float —
correct for a delay, destructive for condition text, where 0 and 0.0 are different
tokens to a Verilog consumer.

Fix

Grammar only; no Python changes.

-equation: (operator | STRING | FLOAT | SCALARCONSTANT)+
+equation: (operator | STRING | EQNUMBER | SCALARCONSTANT)+

-SCALARCONSTANT: /[01]?'[bB][01]/
+SCALARCONSTANT.3: /[01]?'[bB][01]/
+// Numbers inside a condition keep their original text (see equation)
+EQNUMBER.2: /[-+]?[0-9]+(?:\.[0-9]+)?(?:[eE][+-]?[0-9]+)?/

Conditions get their own number terminal, so equation never touches FLOAT and no
conversion callback fires. This works because LALR mode uses a contextual lexer: EQNUMBER
is only reachable inside equation, so delay positions still lex as FLOAT exactly as
before.

Three details that matter if this is edited later:

• SCALARCONSTANT must outrank EQNUMBER (.3 vs .2). Reversed, the number terminal
  takes the leading 1 and 1'b0 splits again.
• EQNUMBER must cover decimals, or A==1.5 splits into 1 and .5.
• EQNUMBER needs [-+]?, or A == -1 fails to parse — FLOAT's -? had been carrying
  that case. This also makes +2 work, which FLOAT never allowed.

Golden files

Two goldens are regenerated. The diff is the corruption being undone:

-  (COND (B == 1.0 'b0 && C == 1.0 'b0 && D == 1.0 'b0)
+  (COND (B == 1'b0 && C == 1'b0 && D == 1'b0)
-  (COND (TE == 0.0 && RB == 1.0 && SB == 1.0)
+  (COND (TE == 0 && RB == 1 && SB == 1)

8 lines in golden/fixpoint.sdf, 8 in golden/spec-example2.sdf. No other golden changes.

Verification

• 346 passed after regenerating the two goldens.
• The unsized form A=='b1 was already correct and is unchanged.
• No file changes accept/reject status: this fixes the text a condition round-trips to,
  not what parses.