Avoid allocating token arrays in Parser for simple declarations

[alanzabihi]

Hypothesis

The parser allocates a tokens array in other() that collects all tokens for a given statement, then passes this array to decl() or rule(). For simple CSS declarations like color: red; or margin: 10px;, this involves:

1. Allocating an array in other()
2. Pushing 3-5 tokens (prop, colon/space, value, semicolon)
3. Passing array to decl()
4. decl() iterates/splices the array

For simple declarations (the vast majority in typical CSS), skip the array entirely. Detect the common pattern inline: word, colon, value tokens, semicolon. Build the Declaration node directly from the token stream:

// In parse(), after getting a word token:
let next = this.tokenizer.nextToken()
if (next[0] === ':') {
  // Fast-path: this is a simple declaration
  let value = this.tokenizer.nextToken() // value token
  // Build Declaration directly without array allocation
}

This eliminates array allocation, push calls, and splice operations for the majority of CSS statements.

Editable surface

• lib/parser.js — add fast-path for simple declarations that avoids token array allocation

What's different from prior work

• Specialize Parser.decl() and Parser.rule() fast paths for common CSS patterns #21 specialized decl/rule fast paths but still used token arrays.
• Reduce Parser token array intermediate allocations in spacesAndCommentsFromEnd #48 optimized spacesAndComments on token arrays (neutral). This eliminates the arrays.
• Flatten Parser.other() token scanning loop to reduce branch depth #31 flattened other() but kept the same allocation pattern.
• The key insight: for simple declarations (no comments, no complex values), we can consume tokens directly from the tokenizer without intermediate collection.

Expected impact

METRIC_A improvement of 2-5ms (large file, thousands of simple declarations). METRIC_B improvement of 2-4ms. Combined 3-5ms.

[alanzabihi]

Polyresearch approval: thesis #74.