encoder: More unexpected EOF handling

Continuing 995fce81 I've decided to rework unexpected EOF testing to run
on truncated versions of all inputs we already have. This uncovered the
following:

- if an individual op decoder returns err=io.EOF - this is unexpected
  and should be turned into io.ErrUnexpectedEOF. If we do not do it
  shows as e.g.

        ogorek_test.go:91: int: no ErrUnexpectedEOF on [:1] truncated stream: v = <nil>  err = &errors.errorString{s:"EOF"}

- there was a potential buffer overrun panic in loadString - the code
  there did not have proper check for line rest length after initial ".
  The panic looks like this:

--- FAIL: TestDecode (0.00s)
panic: runtime error: slice bounds out of range [recovered]
        panic: runtime error: slice bounds out of range

goroutine 7 [running]:
testing.tRunner.func1(0xc4200725b0)
        /home/kirr/src/tools/go/go/src/testing/testing.go:624 +0x267
panic(0x5368c0, 0x603020)
        /home/kirr/src/tools/go/go/src/runtime/panic.go:489 +0x26e
github.com/kisielk/og-rek.(*Decoder).loadString(0xc420063428, 0xc42000e153, 0x0)
        /home/kirr/src/wendelin/neo/g.neo/src/github.com/kisielk/og-rek/ogorek.go:478 +0x22d
github.com/kisielk/og-rek.Decoder.Decode(0xc42012fbc0, 0x6248e8, 0x0, 0x0, 0xc42013ac60, 0x0, 0x0, 0x5f2480, 0xc42000e150)
        /home/kirr/src/wendelin/neo/g.neo/src/github.com/kisielk/og-rek/ogorek.go:166 +0xa83
github.com/kisielk/og-rek.TestDecode(0xc4200725b0)
        /home/kirr/src/wendelin/neo/g.neo/src/github.com/kisielk/og-rek/ogorek_test.go:89 +0x26c7
testing.tRunner(0xc4200725b0, 0x567618)
        /home/kirr/src/tools/go/go/src/testing/testing.go:659 +0x98
created by testing.(*T).Run
        /home/kirr/src/tools/go/go/src/testing/testing.go:701 +0x2df
exit status 2

  and can be triggered by single-character " input.

- we also turn what was previously an "insecure string" into unexpected
  EOF. The reason here is: code could not find " in the end of the
  string. This can be for 2 reasons:

  1. EOF was reached, or
  2. EOL (end of line) was reached

  unfortunately since we are currently using bufio.Buffer.ReadLine()
  which does not indicate whether EOL was there or not and strips it
  from result, we cannot distinguish above cases and properly report
  "insecure string" in 2 case.

  However unexpected EOF even for #2 seems a bit reasonable, so I think
  the change should be ok.

This is explicitly stated in bytes.Buffer documentation and is already
relyed by ogórek in another place:

https://github.com/kisielk/og-rek/blob/c7dbf2e4/ogorek.go#L524

Misc fixes (error + opStop handling)

io.EOF handling was already there in instructions decoding loop, but it
was simply breaking out of the loop and in the end decode
unconditionally was trying to pop top element from stack, and since
stack is empty panicked. E.g. for "" input this way:

--- FAIL: TestDecode (0.00s)
panic: runtime error: index out of range [recovered]
        panic: runtime error: index out of range

goroutine 20 [running]:
testing.tRunner.func1(0xc4200765b0)
        /home/kirr/src/tools/go/go/src/testing/testing.go:624 +0x267
panic(0x5368c0, 0x603010)
        /home/kirr/src/tools/go/go/src/runtime/panic.go:489 +0x26e
github.com/kisielk/og-rek.Decoder.Decode(0xc4200ea5a0, 0x624908, 0x0, 0x0, 0xc4200e6b10, 0x52a180, 0xc4200ee308, 0x0, 0x0)
        /home/kirr/src/wendelin/neo/g.neo/src/github.com/kisielk/og-rek/ogorek.go:250 +0x4bb
github.com/kisielk/og-rek.TestDecode(0xc4200765b0)
        /home/kirr/src/wendelin/neo/g.neo/src/github.com/kisielk/og-rek/ogorek_test.go:80 +0x232f
testing.tRunner(0xc4200765b0, 0x567648)
        /home/kirr/src/tools/go/go/src/testing/testing.go:659 +0x98
created by testing.(*T).Run
        /home/kirr/src/tools/go/go/src/testing/testing.go:701 +0x2df
exit status 2

Fix it via reworking EOF handling:

- if we see EOF and no instructions was decoded so far - just return it
  to the caller

- if we see EOF in the middle of the stream - as valid pickle streams
  end with Stop opcode - it is unexpected EOF - return it as such.

A couple of tests had to be adjusted in TestDecode - as input there was
not a valid pickle - not ending with "." (which python pickle decoder
also does not decode at all).

There was a break already on a `case opStop:` but break inside switch
only breaks out of this switch, not outer construct - looks like it was
a thinko.

So on opStop let's explicitly break out of instruction processing loop.

If we do not do so on e.g "I5\n.I7\n.N." input first call to .Decode()
returns None (from N.), not 5 and second call to Decode() panics (for
this see next patch).

Previously errors from underlying .encode() were ignored completely and
not returned to callers.

Implement opMemoize/opShortBinUnicode/opFrame from 'Protocol 4'

Merge branch 'feature/FixReadLineBug' of https://github.com/CodeRushing/og-rek into CodeRushing-feature/FixReadLineBug

fix when the first char is \, it cause UnquoteChar to return syntax e…

fix for buffer limit in loading string

Handle write and read errors.

This makes it simpler to handle errors in code importing this package.

Implement r opcode LONG_BINPUT

Fix struct encoding

There was accidentally an extra empty dict.  Also, reduce the scope of a
local variable.

Changed the filenames ogórek.go and ogórek_test.go to use the precomposed Latin Small Letter O With Acute.