Fix comment parsing to respect word boundaries around '#'

[david-littlefarmer]

Summary

• parseComment used a naive strings.SplitN(s, "#", 2), treating any # in a line as a comment marker. webrpc's real ridl lexer only starts a comment when # is not touching an in-progress word token — e.g. in v0.0.1#beta, the # is part of the value, not a comment. ridlfmt was silently altering parsed values by inserting a space before such a #.
• reduceSpaces in processor.go had the same naive-hash-index bug and fed corrupted (space-squashed) input into parseComment, so it's fixed the same way.
• Updated the two pre-existing pathological example fixtures (_examples/e1.ridl, main_test.go) that had a # glued directly to a value with no separating space — that's genuinely ambiguous per webrpc's grammar, so a space was added to make them valid, unambiguous RIDL.
• Added formatter/comments_test.go covering the exact cases from the issue.

Fixes #17

Examples

Case 1 — # glued to a value (version=v0.0.1#beta)

	Output
Input	version=v0.0.1#beta
Before (bug)	version = v0.0.1 # beta — split the value, changing it from v0.0.1#beta to v0.0.1
After (fixed)	version = v0.0.1#beta — value preserved intact

Case 2 — trailing # with nothing after it (name =foo-bar#)

	Output
Input	name =foo-bar#
Before (bug)	name = foo-bar # — turned the trailing # into an (empty) comment, changing the value from foo-bar# to foo-bar
After (fixed)	name = foo-bar# — value preserved intact

A real inline comment (# after whitespace) is unaffected in both versions, e.g. version = v1 # my comment stays version = v1 # my comment.

Test plan

• go build ./...
• go test ./... (all pass, including new formatter package tests)
• gofmt -l . / go vet ./... clean
• Manually verified before/after behavior above against master and this branch