Fix v0.1.3 regression: stop coercing stored string values

[gillisd]

What broke

v0.1.3 rewrote Vivification#vivify_value from a class-based case/when into a pattern-matching case/in that runs every stored value through INTEGER / FLOAT / DATELIKE regexes:

def vivify_value(value)
  case value
  in Hash then self.class.new(value, root: false)
  in Array then vivify_array(value)
  in INTEGER then value.to_i
  in FLOAT then value.to_f
  in DATELIKE
    require "datetime"
    DateTime.parse(it)
  else value
  end
end

vivify_value runs on every construction (#initialize → #vivify) and every assignment (#[]=), so this silently corrupts any string containing a digit:

Input	v0.1.2	v0.1.3
Flexor.new(address: "123 Main Street").address	"123 Main Street"	123
Flexor.new(note: "call me at 3pm").note	"call me at 3pm"	0.0
Flexor.new(phone: "555-1234").phone	"555-1234"	555
Flexor.new(ts: "2023-01-15T10:30:00").ts	"2023-01-15T10:30:00"	2023

The branch was internally broken too — the DATELIKE arm does require "datetime" (no such file; it's date) and calls it outside a block — but INTEGER's [0-9-]+ matches date strings first, so that arm was dead code masking two more bugs.

Why no test caught it

Every string value in the suite is digit-free ("alice", "deep", "localhost", "NYC", …), and every digit-bearing datum is either already an Integer (age: 30, id: 1) or nested inside an Array — and vivify_array was never changed, so array elements skip the coercion entirely. The coercion code path simply had no exercising test.

(The new top-level patterns_spec.rb added in the same changeset lived outside spec/, so rake spec never ran it, and it contained only a skip.)

Fix

• Restore vivify_value to pass non-Hash/Array values through unchanged (matching the sibling vivify_array).
• Add regression coverage for construction and assignment of number-/date-like strings (8 specs that fail on v0.1.3, pass after the fix).
• Remove the stray root-level patterns_spec.rb debris from the abandoned coercion experiment.
• Bump the version to 0.1.4 so the fix is releasable (0.1.3 should be considered broken).

Notes

• The other v0.1.3 change — refactoring Serialization#init_with to a .then { it[...] } pipeline — is not a real bug on the supported platform. it refers to the block argument from Ruby 3.4 onward, and the gemspec requires >= 3.4. It's functionally identical to v0.1.2 there.
• This regression was the author's attempt to address the open issues.rec entry "'Coerce' error when using sum". That issue (Flexor not implementing the numeric coerce protocol) is separate and still open — it should be fixed without mutating stored data. Worth a follow-up.